Authors: José M. Alcalde-Llergo, Enrique Yeguas-Bolívar, Pilar Aparicio-Martínez, Andrea Zingoni, Juri Taborri, Sara Pinzi
Dyslexia is a neurodevelopmental disorder that is estimated to affect about 5-10% of the population. In particular, phonological dyslexia causes problems in connecting the sounds of words with their written forms. This results in difficulties such as slow reading speed, inaccurate reading, and difficulty decoding unfamiliar words. Moreover, dyslexia can also be a challenging and frustrating experience for students as they may feel misunderstood or stigmatized by their peers or educators. For these reasons, the use of compensatory tools and strategies is of crucial importance for dyslexic students to have the same opportunities as non-dyslexic ones. However, generally, people underestimate the problem and are not aware of the importance of support methodologies. In the light of this, the main purpose of this paper is to propose a virtual reality (VR) serious game through which teachers, students and, in general, non-dyslexic people could understand which are some of the issues of student with dyslexia and the fundamental utility of offering support to them. In the game, players must create a potion by following a recipe written in an alphabet that is specifically designed to replicate the reading difficulties experienced by individuals with dyslexia. The task must be solved first without any help and then by receiving supporting tools and strategies with the idea that the player can put himself in the place of the dyslexic person and understand the real need for support methodologies.
Authors: Xiaolong Huang, Qiankun Li, Xueran Li, Xuesong Gao
Visual fine-tuning has garnered significant attention with the rise of pre-trained vision models. The current prevailing method, full fine-tuning, suffers from the issue of knowledge forgetting as it focuses solely on fitting the downstream training set. In this paper, we propose a novel weight rollback-based fine-tuning method called OLOR (One step Learning, One step Review). OLOR combines fine-tuning with optimizers, incorporating a weight rollback term into the weight update term at each step. This ensures consistency in the weight range of upstream and downstream models, effectively mitigating knowledge forgetting and enhancing fine-tuning performance. In addition, a layer-wise penalty is presented to employ penalty decay and the diversified decay rate to adjust the weight rollback levels of layers for adapting varying downstream tasks. Through extensive experiments on various tasks such as image classification, object detection, semantic segmentation, and instance segmentation, we demonstrate the general applicability and state-of-the-art performance of our proposed OLOR. Code is available at https://github.com/rainbow-xiao/OLOR-AAAI-2024.
Authors: Danzel Serrano, Jakub Szymkowiak, Przemyslaw Musialski
Recently proposed methods for implicitly representing signals such as images, scenes, or geometries using coordinate-based neural network architectures often do not leverage the choice of activation functions, or do so only to a limited extent. In this paper, we introduce the Hyperbolic Oscillation function (HOSC), a novel activation function with a controllable sharpness parameter. Unlike any previous activations, HOSC has been specifically designed to better capture sudden changes in the input signal, and hence sharp or acute features of the underlying data, as well as smooth low-frequency transitions. Due to its simplicity and modularity, HOSC offers a plug-and-play functionality that can be easily incorporated into any existing method employing a neural network as a way of implicitly representing a signal. We benchmark HOSC against other popular activations in an array of general tasks, empirically showing an improvement in the quality of obtained representations, provide the mathematical motivation behind the efficacy of HOSC, and discuss its limitations.
Authors: Chaitanya Patel, Shaojie Bai, Te-Li Wang, Jason Saragih, Shih-En Wei
Virtual Reality (VR) bares promise of social interactions that can feel more immersive than other media. Key to this is the ability to accurately animate a photorealistic avatar of one's likeness while wearing a VR headset. Although high quality registration of person-specific avatars to headset-mounted camera (HMC) images is possible in an offline setting, the performance of generic realtime models are significantly degraded. Online registration is also challenging due to oblique camera views and differences in modality. In this work, we first show that the domain gap between the avatar and headset-camera images is one of the primary sources of difficulty, where a transformer-based architecture achieves high accuracy on domain-consistent data, but degrades when the domain-gap is re-introduced. Building on this finding, we develop a system design that decouples the problem into two parts: 1) an iterative refinement module that takes in-domain inputs, and 2) a generic avatar-guided image-to-image style transfer module that is conditioned on current estimation of expression and head pose. These two modules reinforce each other, as image style transfer becomes easier when close-to-ground-truth examples are shown, and better domain-gap removal helps registration. Our system produces high-quality results efficiently, obviating the need for costly offline registration to generate personalized labels. We validate the accuracy and efficiency of our approach through extensive experiments on a commodity headset, demonstrating significant improvements over direct regression methods as well as offline registration.
Authors: Michael Gerstenberger, Dominic Juestel, Silviu Bodea
During deep sleep and under anaesthesia spontaneous patterns of cortical activation frequently take the form of slow travelling waves. Slow wave sleep is an important cognitive state especially because of its relevance for memory consolidation. However, despite extensive research the exact mechanisms are still ill-understood. Novel methods such as high speed widefield imaging of GCamP activity offer new potentials. Here we show how data recorded from transgenic mice under anesthesia can be processed to analyze sources, sinks and patterns of flow. To make the best possible use of the data novel means of data processing are necessary. Therefore, we (1) give a an brief account on processes that play a role in generating slow waves and demonstrate (2) a novel approach to characterize its patterns in GCamP recordings. While slow waves are highly variable, it shows that some are surprisingly similar. To enable quantitative means of analysis and examine the structure of such prototypical events we propose a novel approach for the characterization of slow waves: The Helmholtz-Decomposition of gradient-based Dense Optical Flow of the pixeldense GCamP contrast (df/f). It allows to detect the sources and sinks of activation and discern them from global patterns of neural flow. Aggregated features can be analyzed with variational autoencoders. The results unravel regularities between slow waves and shows how they relate to the experimental conditions. The approach reveals a complex topology of different features in latent slow wave space and identifies prototypical examples for each stage.
Authors: Mazal Bethany, Brandon Wherry, Nishant Vishwamitra, Peyman Najafirad
Social media platforms are being increasingly used by malicious actors to share unsafe content, such as images depicting sexual activity, cyberbullying, and self-harm. Consequently, major platforms use artificial intelligence (AI) and human moderation to obfuscate such images to make them safer. Two critical needs for obfuscating unsafe images is that an accurate rationale for obfuscating image regions must be provided, and the sensitive regions should be obfuscated (\textit{e.g.} blurring) for users' safety. This process involves addressing two key problems: (1) the reason for obfuscating unsafe images demands the platform to provide an accurate rationale that must be grounded in unsafe image-specific attributes, and (2) the unsafe regions in the image must be minimally obfuscated while still depicting the safe regions. In this work, we address these key issues by first performing visual reasoning by designing a visual reasoning model (VLM) conditioned on pre-trained unsafe image classifiers to provide an accurate rationale grounded in unsafe image attributes, and then proposing a counterfactual explanation algorithm that minimally identifies and obfuscates unsafe regions for safe viewing, by first utilizing an unsafe image classifier attribution matrix to guide segmentation for a more optimal subregion segmentation followed by an informed greedy search to determine the minimum number of subregions required to modify the classifier's output based on attribution score. Extensive experiments on uncurated data from social networks emphasize the efficacy of our proposed method. We make our code available at: https://github.com/SecureAIAutonomyLab/ConditionalVLM
Authors: Oliver Limoyo, Jimmy Li, Dmitriy Rivkin, Jonathan Kelly, Gregory Dudek
We introduce PhotoBot, a framework for automated photo acquisition based on an interplay between high-level human language guidance and a robot photographer. We propose to communicate photography suggestions to the user via a reference picture that is retrieved from a curated gallery. We exploit a visual language model (VLM) and an object detector to characterize reference pictures via textual descriptions and use a large language model (LLM) to retrieve relevant reference pictures based on a user's language query through text-based reasoning. To correspond the reference picture and the observed scene, we exploit pre-trained features from a vision transformer capable of capturing semantic similarity across significantly varying images. Using these features, we compute pose adjustments for an RGB-D camera by solving a Perspective-n-Point (PnP) problem. We demonstrate our approach on a real-world manipulator equipped with a wrist camera. Our user studies show that photos taken by PhotoBot are often more aesthetically pleasing than those taken by users themselves, as measured by human feedback.
Authors: Md Zahangir Alom, Quynh T. Tran, Brent A. Orr
Histologic examination plays a crucial role in oncology research and diagnostics. The adoption of digital scanning of whole slide images (WSI) has created an opportunity to leverage deep learning-based image classification methods to enhance diagnosis and risk stratification. Technical limitations of current approaches to training deep convolutional neural networks (DCNN) result in suboptimal model performance and make training and deployment of comprehensive classification models unobtainable. In this study, we introduce a novel approach that addresses the main limitations of traditional histopathology classification model training. Our method, termed Learned Resizing with Efficient Training (LRET), couples efficient training techniques with image resizing to facilitate seamless integration of larger histology image patches into state-of-the-art classification models while preserving important structural information.
We used the LRET method coupled with two distinct resizing techniques to train three diverse histology image datasets using multiple diverse DCNN architectures. Our findings demonstrate a significant enhancement in classification performance and training efficiency. Across the spectrum of experiments, LRET consistently outperforms existing methods, yielding a substantial improvement of 15-28% in accuracy for a large-scale, multiclass tumor classification task consisting of 74 distinct brain tumor types. LRET not only elevates classification accuracy but also substantially reduces training times, unlocking the potential for faster model development and iteration. The implications of this work extend to broader applications within medical imaging and beyond, where efficient integration of high-resolution images into deep learning pipelines is paramount for driving advancements in research and clinical practice.
Authors: Ka-Hei Hui, Aditya Sanghi, Arianna Rampini, Kamal Rahimi Malekshan, Zhengzhe Liu, Hooman Shayani, Chi-Wing Fu
Significant progress has been made in training large generative models for natural language and images. Yet, the advancement of 3D generative models is hindered by their substantial resource demands for training, along with inefficient, non-compact, and less expressive representations. This paper introduces Make-A-Shape, a new 3D generative model designed for efficient training on a vast scale, capable of utilizing 10 millions publicly-available shapes. Technical-wise, we first innovate a wavelet-tree representation to compactly encode shapes by formulating the subband coefficient filtering scheme to efficiently exploit coefficient relations. We then make the representation generatable by a diffusion model by devising the subband coefficients packing scheme to layout the representation in a low-resolution grid. Further, we derive the subband adaptive training strategy to train our model to effectively learn to generate coarse and detail wavelet coefficients. Last, we extend our framework to be controlled by additional input conditions to enable it to generate shapes from assorted modalities, e.g., single/multi-view images, point clouds, and low-resolution voxels. In our extensive set of experiments, we demonstrate various applications, such as unconditional generation, shape completion, and conditional generation on a wide range of modalities. Our approach not only surpasses the state of the art in delivering high-quality results but also efficiently generates shapes within a few seconds, often achieving this in just 2 seconds for most conditions.
Authors: Mingyuan Zhou, Rakib Hyder, Ziwei Xuan, Guojun Qi
Recent advances in 3D avatar generation have gained significant attentions. These breakthroughs aim to produce more realistic animatable avatars, narrowing the gap between virtual and real-world experiences. Most of existing works employ Score Distillation Sampling (SDS) loss, combined with a differentiable renderer and text condition, to guide a diffusion model in generating 3D avatars. However, SDS often generates oversmoothed results with few facial details, thereby lacking the diversity compared with ancestral sampling. On the other hand, other works generate 3D avatar from a single image, where the challenges of unwanted lighting effects, perspective views, and inferior image quality make them difficult to reliably reconstruct the 3D face meshes with the aligned complete textures. In this paper, we propose a novel 3D avatar generation approach termed UltrAvatar with enhanced fidelity of geometry, and superior quality of physically based rendering (PBR) textures without unwanted lighting. To this end, the proposed approach presents a diffuse color extraction model and an authenticity guided texture diffusion model. The former removes the unwanted lighting effects to reveal true diffuse colors so that the generated avatars can be rendered under various lighting conditions. The latter follows two gradient-based guidances for generating PBR textures to render diverse face-identity features and details better aligning with 3D mesh geometry. We demonstrate the effectiveness and robustness of the proposed method, outperforming the state-of-the-art methods by a large margin in the experiments.
Authors: Yuefang Gao, Yuhao Xie, Zeke Zexi Hu, Tianshui Chen, Liang Lin
Domain shift poses a significant challenge in Cross-Domain Facial Expression Recognition (CD-FER) due to the distribution variation across different domains. Current works mainly focus on learning domain-invariant features through global feature adaptation, while neglecting the transferability of local features. Additionally, these methods lack discriminative supervision during training on target datasets, resulting in deteriorated feature representation in target domain. To address these limitations, we propose an Adaptive Global-Local Representation Learning and Selection (AGLRLS) framework. The framework incorporates global-local adversarial adaptation and semantic-aware pseudo label generation to enhance the learning of domain-invariant and discriminative feature during training. Meanwhile, a global-local prediction consistency learning is introduced to improve classification results during inference. Specifically, the framework consists of separate global-local adversarial learning modules that learn domain-invariant global and local features independently. We also design a semantic-aware pseudo label generation module, which computes semantic labels based on global and local features. Moreover, a novel dynamic threshold strategy is employed to learn the optimal thresholds by leveraging independent prediction of global and local features, ensuring filtering out the unreliable pseudo labels while retaining reliable ones. These labels are utilized for model optimization through the adversarial learning process in an end-to-end manner. During inference, a global-local prediction consistency module is developed to automatically learn an optimal result from multiple predictions. We conduct comprehensive experiments and analysis based on a fair evaluation benchmark. The results demonstrate that the proposed framework outperforms the current competing methods by a substantial margin.
Unsupervised video object learning seeks to decompose video scenes into structural object representations without any supervision from depth, optical flow, or segmentation. We present VONet, an innovative approach that is inspired by MONet. While utilizing a U-Net architecture, VONet employs an efficient and effective parallel attention inference process, generating attention masks for all slots simultaneously. Additionally, to enhance the temporal consistency of each mask across consecutive video frames, VONet develops an object-wise sequential VAE framework. The integration of these innovative encoder-side techniques, in conjunction with an expressive transformer-based decoder, establishes VONet as the leading unsupervised method for object learning across five MOVI datasets, encompassing videos of diverse complexities. Code is available at https://github.com/hnyu/vonet.
Authors: Kuan-Ting Kuo, Dana Moukheiber, Sebastian Cajas Ordonez, David Restrepo, Atika Rahman Paddo, Tsung-Yu Chen, Lama Moukheiber, Mira Moukheiber, Sulaiman Moukheiber, Saptarshi Purkayastha, Po-Chih Kuo, Leo Anthony Celi
Dengue fever presents a substantial challenge in developing countries where sanitation infrastructure is inadequate. The absence of comprehensive healthcare systems exacerbates the severity of dengue infections, potentially leading to life-threatening circumstances. Rapid response to dengue outbreaks is also challenging due to limited information exchange and integration. While timely dengue outbreak forecasts have the potential to prevent such outbreaks, the majority of dengue prediction studies have predominantly relied on data that impose significant burdens on individual countries for collection. In this study, our aim is to improve health equity in resource-constrained countries by exploring the effectiveness of high-resolution satellite imagery as a nontraditional and readily accessible data source. By leveraging the wealth of publicly available and easily obtainable satellite imagery, we present a scalable satellite extraction framework based on Sentinel Hub, a cloud-based computing platform. Furthermore, we introduce DengueNet, an innovative architecture that combines Vision Transformer, Radiomics, and Long Short-term Memory to extract and integrate spatiotemporal features from satellite images. This enables dengue predictions on an epi-week basis. To evaluate the effectiveness of our proposed method, we conducted experiments on five municipalities in Colombia. We utilized a dataset comprising 780 high-resolution Sentinel-2 satellite images for training and evaluation. The performance of DengueNet was assessed using the mean absolute error (MAE) metric. Across the five municipalities, DengueNet achieved an average MAE of 43.92. Our findings strongly support the efficacy of satellite imagery as a valuable resource for dengue prediction, particularly in informing public health policies within countries where manually collected data is scarce and dengue virus prevalence is severe.
Authors: Nhat M. Hoang, Kehong Gong, Chuan Guo, Michael Bi Mi
Controllable generation of 3D human motions becomes an important topic as the world embraces digital transformation. Existing works, though making promising progress with the advent of diffusion models, heavily rely on meticulously captured and annotated (e.g., text) high-quality motion corpus, a resource-intensive endeavor in the real world. This motivates our proposed MotionMix, a simple yet effective weakly-supervised diffusion model that leverages both noisy and unannotated motion sequences. Specifically, we separate the denoising objectives of a diffusion model into two stages: obtaining conditional rough motion approximations in the initial $T-T^*$ steps by learning the noisy annotated motions, followed by the unconditional refinement of these preliminary motions during the last $T^*$ steps using unannotated motions. Notably, though learning from two sources of imperfect data, our model does not compromise motion generation quality compared to fully supervised approaches that access gold data. Extensive experiments on several benchmarks demonstrate that our MotionMix, as a versatile framework, consistently achieves state-of-the-art performances on text-to-motion, action-to-motion, and music-to-dance tasks.
Authors: Tao Chen, Yazhou Yao, Xingguo Huang, Zechao Li, Liqiang Nie, Jinhui Tang
The image-level label has prevailed in weakly supervised semantic segmentation tasks due to its easy availability. Since image-level labels can only indicate the existence or absence of specific categories of objects, visualization-based techniques have been widely adopted to provide object location clues. Considering class activation maps (CAMs) can only locate the most discriminative part of objects, recent approaches usually adopt an expansion strategy to enlarge the activation area for more integral object localization. However, without proper constraints, the expanded activation will easily intrude into the background region. In this paper, we propose spatial structure constraints (SSC) for weakly supervised semantic segmentation to alleviate the unwanted object over-activation of attention expansion. Specifically, we propose a CAM-driven reconstruction module to directly reconstruct the input image from deep CAM features, which constrains the diffusion of last-layer object attention by preserving the coarse spatial structure of the image content. Moreover, we propose an activation self-modulation module to refine CAMs with finer spatial structure details by enhancing regional consistency. Without external saliency models to provide background clues, our approach achieves 72.7\% and 47.0\% mIoU on the PASCAL VOC 2012 and COCO datasets, respectively, demonstrating the superiority of our proposed approach.
Authors: Haoxiang Yang, Chengguo Yuan, Yabin Zhu, Lan Chen, Xiao Wang, Jin Tang
The mainstream human activity recognition (HAR) algorithms are developed based on RGB cameras, which are easily influenced by low-quality images (e.g., low illumination, motion blur). Meanwhile, the privacy protection issue caused by ultra-high definition (HD) RGB cameras aroused more and more people's attention. Inspired by the success of event cameras which perform better on high dynamic range, no motion blur, and low energy consumption, we propose to recognize human actions based on the event stream. We propose a lightweight uncertainty-aware information propagation based Mobile-Former network for efficient pattern recognition, which aggregates the MobileNet and Transformer network effectively. Specifically, we first embed the event images using a stem network into feature representations, then, feed them into uncertainty-aware Mobile-Former blocks for local and global feature learning and fusion. Finally, the features from MobileNet and Transformer branches are concatenated for pattern recognition. Extensive experiments on multiple event-based recognition datasets fully validated the effectiveness of our model. The source code of this work will be released at https://github.com/Event-AHU/Uncertainty_aware_MobileFormer.
Authors: Dimitrios Sinodinos, Narges Armanfard
Multitask learning (MTL) has gained prominence for its ability to jointly predict multiple tasks, achieving better per-task performance while using fewer per-task model parameters than single-task learning. More recently, decoder-focused architectures have considerably improved multitask performance by refining task predictions using the features of other related tasks. However, most of these refinement methods fail to simultaneously capture local and global task-specific representations, as well as cross-task patterns in a parameter-efficient manner. In this paper, we introduce the Efficient Multitask Affinity Learning Network (EMA-Net), which is a lightweight framework that enhances the task refinement capabilities of multitask networks. EMA-Net adeptly captures local, global, and cross-task interactions using our novel Cross-Task Affinity Learning (CTAL) module. The key innovation of CTAL lies in its ability to manipulate task affinity matrices in a manner that is optimally suited to apply parameter-efficient grouped convolutions without worrying about information loss. Our results show that we achieve state-of-the-art MTL performance for CNN-based decoder-focused models while using substantially fewer model parameters. Our code is publicly available at https://github.com/Armanfard-Lab/EMA-Net.
Authors: Yuantao Yin, Ping Yin
Few-shot object detection(FSOD) aims to design methods to adapt object detectors efficiently with only few annotated samples. Fine-tuning has been shown to be an effective and practical approach. However, previous works often take the classical base-novel two stage fine-tuning procedure but ignore the implicit stability-plasticity contradiction among different modules. Specifically, the random re-initialized classifiers need more plasticity to adapt to novel samples. The other modules inheriting pre-trained weights demand more stability to reserve their class-agnostic knowledge. Regular fine-tuning which couples the optimization of these two parts hurts the model generalization in FSOD scenarios. In this paper, we find that this problem is prominent in the end-to-end object detector Sparse R-CNN for its multi-classifier cascaded architecture. We propose to mitigate this contradiction by a new three-stage fine-tuning procedure by introducing an addtional plasticity classifier fine-tuning(PCF) stage. We further design the multi-source ensemble(ME) technique to enhance the generalization of the model in the final fine-tuning stage. Extensive experiments verify that our method is effective in regularizing Sparse R-CNN, outperforming previous methods in the FSOD benchmark.
Authors: Georgios Ioannides, Aman Chadha, Aaron Elkins
We propose the Multi-Head Gaussian Adaptive Attention Mechanism (GAAM), a novel probabilistic attention framework, and the Gaussian Adaptive Transformer (GAT), designed to enhance information aggregation across multiple modalities, including Speech, Text and Vision. GAAM integrates learnable mean and variance into its attention mechanism, implemented in a Multi-Headed framework enabling it to collectively model any Probability Distribution for dynamic recalibration of feature significance. This method demonstrates significant improvements, especially with highly non-stationary data, surpassing the state-of-the-art attention techniques in model performance (up to approximately +20% in accuracy) by identifying key elements within the feature space. GAAM's compatibility with dot-product-based attention models and relatively low number of parameters showcases its adaptability and potential to boost existing attention frameworks. Empirically, GAAM exhibits superior adaptability and efficacy across a diverse range of tasks, including emotion recognition in speech, image classification, and text classification, thereby establishing its robustness and versatility in handling multi-modal data. Furthermore, we introduce the Importance Factor (IF), a new learning-based metric that enhances the explainability of models trained with GAAM-based methods. Overall, GAAM represents an advancement towards development of better performing and more explainable attention models across multiple modalities.
Authors: Junxiao Shen, Matthias De Lange, Xuhai "Orson" Xu, Enmin Zhou, Ran Tan, Naveen Suda, Maciej Lazarewicz, Per Ola Kristensson, Amy Karlson, Evan Strasnick
Static machine learning methods in gesture recognition assume that training and test data come from the same underlying distribution. However, in real-world applications involving gesture recognition on wrist-worn devices, data distribution may change over time. We formulate this problem of adapting recognition models to new tasks, where new data patterns emerge, as open-world gesture recognition (OWGR). We propose leveraging continual learning to make machine learning models adaptive to new tasks without degrading performance on previously learned tasks. However, the exploration of parameters for questions around when and how to train and deploy recognition models requires time-consuming user studies and is sometimes impractical. To address this challenge, we propose a design engineering approach that enables offline analysis on a collected large-scale dataset with various parameters and compares different continual learning methods. Finally, design guidelines are provided to enhance the development of an open-world wrist-worn gesture recognition process.
Authors: Junxiao Shen, Xuhai Xu, Ran Tan, Amy Karlson, Evan Strasnick
Training a real-time gesture recognition model heavily relies on annotated data. However, manual data annotation is costly and demands substantial human effort. In order to address this challenge, we propose a novel annotation model that can automatically annotate gesture classes and identify their temporal ranges. Our ablation study demonstrates that our annotation model design surpasses the baseline in terms of both gesture classification accuracy (3-4\% improvement) and localization accuracy (71-75\% improvement). We believe that this annotation model has immense potential to improve the training of downstream gesture recognition models using unlabeled datasets.
Authors: Kuofeng Gao, Yang Bai, Jindong Gu, Shu-Tao Xia, Philip Torr, Zhifeng Li, Wei Liu
Large vision-language models (VLMs) such as GPT-4 have achieved exceptional performance across various multi-modal tasks. However, the deployment of VLMs necessitates substantial energy consumption and computational resources. Once attackers maliciously induce high energy consumption and latency time (energy-latency cost) during inference of VLMs, it will exhaust computational resources. In this paper, we explore this attack surface about availability of VLMs and aim to induce high energy-latency cost during inference of VLMs. We find that high energy-latency cost during inference of VLMs can be manipulated by maximizing the length of generated sequences. To this end, we propose verbose images, with the goal of crafting an imperceptible perturbation to induce VLMs to generate long sentences during inference. Concretely, we design three loss objectives. First, a loss is proposed to delay the occurrence of end-of-sequence (EOS) token, where EOS token is a signal for VLMs to stop generating further tokens. Moreover, an uncertainty loss and a token diversity loss are proposed to increase the uncertainty over each generated token and the diversity among all tokens of the whole generated sequence, respectively, which can break output dependency at token-level and sequence-level. Furthermore, a temporal weight adjustment algorithm is proposed, which can effectively balance these losses. Extensive experiments demonstrate that our verbose images can increase the length of generated sequences by 7.87 times and 8.56 times compared to original images on MS-COCO and ImageNet datasets, which presents potential challenges for various applications. Our code is available at https://github.com/KuofengGao/Verbose_Images.
Authors: Nicolás Ayobi, Santiago Rodríguez, Alejandra Pérez, Isabela Hernández, Nicolás Aparicio, Eugénie Dessevres, Sebastián Peña, Jessica Santander, Juan Ignacio Caicedo, Nicolás Fernández, Pablo Arbeláez
This paper presents the Holistic and Multi-Granular Surgical Scene Understanding of Prostatectomies (GraSP) dataset, a curated benchmark that models surgical scene understanding as a hierarchy of complementary tasks with varying levels of granularity. Our approach enables a multi-level comprehension of surgical activities, encompassing long-term tasks such as surgical phases and steps recognition and short-term tasks including surgical instrument segmentation and atomic visual actions detection. To exploit our proposed benchmark, we introduce the Transformers for Actions, Phases, Steps, and Instrument Segmentation (TAPIS) model, a general architecture that combines a global video feature extractor with localized region proposals from an instrument segmentation model to tackle the multi-granularity of our benchmark. Through extensive experimentation, we demonstrate the impact of including segmentation annotations in short-term recognition tasks, highlight the varying granularity requirements of each task, and establish TAPIS's superiority over previously proposed baselines and conventional CNN-based models. Additionally, we validate the robustness of our method across multiple public benchmarks, confirming the reliability and applicability of our dataset. This work represents a significant step forward in Endoscopic Vision, offering a novel and comprehensive framework for future research towards a holistic understanding of surgical procedures.
Authors: Jiahao Nie, Zhiwei He, Xudong Lv, Xueyi Zhou, Dong-Kyu Chae, Fei Xie
Category-specific models are provenly valuable methods in 3D single object tracking (SOT) regardless of Siamese or motion-centric paradigms. However, such over-specialized model designs incur redundant parameters, thus limiting the broader applicability of 3D SOT task. This paper first introduces unified models that can simultaneously track objects across all categories using a single network with shared model parameters. Specifically, we propose to explicitly encode distinct attributes associated to different object categories, enabling the model to adapt to cross-category data. We find that the attribute variances of point cloud objects primarily occur from the varying size and shape (e.g., large and square vehicles v.s. small and slender humans). Based on this observation, we design a novel point set representation learning network inheriting transformer architecture, termed AdaFormer, which adaptively encodes the dynamically varying shape and size information from cross-category data in a unified manner. We further incorporate the size and shape prior derived from the known template targets into the model's inputs and learning objective, facilitating the learning of unified representation. Equipped with such designs, we construct two category-unified models SiamCUT and MoCUT.Extensive experiments demonstrate that SiamCUT and MoCUT exhibit strong generalization and training stability. Furthermore, our category-unified models outperform the category-specific counterparts by a significant margin (e.g., on KITTI dataset, 12% and 3% performance gains on the Siamese and motion paradigms). Our code will be available.
Authors: Zhongxuan Wang, Leo Xu
The nature of deep neural networks has given rise to a variety of attacks, but little work has been done to address the effect of adversarial attacks on segmentation models trained on MRI datasets. In light of the grave consequences that such attacks could cause, we explore four models from the U-Net family and examine their responses to the Fast Gradient Sign Method (FGSM) attack. We conduct FGSM attacks on each of them and experiment with various schemes to conduct the attacks. In this paper, we find that medical imaging segmentation models are indeed vulnerable to adversarial attacks and that there is a negligible correlation between parameter size and adversarial attack success. Furthermore, we show that using a different loss function than the one used for training yields higher adversarial attack success, contrary to what the FGSM authors suggested. In future efforts, we will conduct the experiments detailed in this paper with more segmentation models and different attacks. We will also attempt to find ways to counteract the attacks by using model ensembles or special data augmentations. Our code is available at https://github.com/ZhongxuanWang/adv_attk
Authors: Yinchao Ma, Yuyang Tang, Wenfei Yang, Tianzhu Zhang, Jinpeng Zhang, Mengxue Kang
Single object tracking aims to locate the target object in a video sequence according to the state specified by different modal references, including the initial bounding box (BBOX), natural language (NL), or both (NL+BBOX). Due to the gap between different modalities, most existing trackers are designed for single or partial of these reference settings and overspecialize on the specific modality. Differently, we present a unified tracker called UVLTrack, which can simultaneously handle all three reference settings (BBOX, NL, NL+BBOX) with the same parameters. The proposed UVLTrack enjoys several merits. First, we design a modality-unified feature extractor for joint visual and language feature learning and propose a multi-modal contrastive loss to align the visual and language features into a unified semantic space. Second, a modality-adaptive box head is proposed, which makes full use of the target reference to mine ever-changing scenario features dynamically from video contexts and distinguish the target in a contrastive way, enabling robust performance in different reference settings. Extensive experimental results demonstrate that UVLTrack achieves promising performance on seven visual tracking datasets, three vision-language tracking datasets, and three visual grounding datasets. Codes and models will be open-sourced at https://github.com/OpenSpaceAI/UVLTrack.
Authors: Yanlong Zang, Han Yang, Jiaxu Miao, Yi Yang
Image-based virtual try-on systems,which fit new garments onto human portraits,are gaining research attention.An ideal pipeline should preserve the static features of clothes(like textures and logos)while also generating dynamic elements(e.g.shadows,folds)that adapt to the model's pose and environment.Previous works fail specifically in generating dynamic features,as they preserve the warped in-shop clothes trivially with predicted an alpha mask by composition.To break the dilemma of over-preserving and textures losses,we propose a novel diffusion-based Product-level virtual try-on pipeline,\ie PLTON, which can preserve the fine details of logos and embroideries while producing realistic clothes shading and wrinkles.The main insights are in three folds:1)Adaptive Dynamic Rendering:We take a pre-trained diffusion model as a generative prior and tame it with image features,training a dynamic extractor from scratch to generate dynamic tokens that preserve high-fidelity semantic information. Due to the strong generative power of the diffusion prior,we can generate realistic clothes shadows and wrinkles.2)Static Characteristics Transformation: High-frequency Map(HF-Map)is our fundamental insight for static representation.PLTON first warps in-shop clothes to the target model pose by a traditional warping network,and uses a high-pass filter to extract an HF-Map for preserving static cloth features.The HF-Map is used to generate modulation maps through our static extractor,which are injected into a fixed U-net to synthesize the final result.To enhance retention,a Two-stage Blended Denoising method is proposed to guide the diffusion process for correct spatial layout and color.PLTON is finetuned only with our collected small-size try-on dataset.Extensive quantitative and qualitative experiments on 1024 768 datasets demonstrate the superiority of our framework in mimicking real clothes dynamics.
Authors: Navin Ranjan, Andreas Savakis
Vision transformers (ViTs) have demonstrated remarkable performance across various visual tasks. However, ViT models suffer from substantial computational and memory requirements, making it challenging to deploy them on resource-constrained platforms. Quantization is a popular approach for reducing model size, but most studies mainly focus on equal bit-width quantization for the entire network, resulting in sub-optimal solutions. While there are few works on mixed precision quantization (MPQ) for ViTs, they typically rely on search space-based methods or employ mixed precision arbitrarily. In this paper, we introduce LRP-QViT, an explainability-based method for assigning mixed-precision bit allocations to different layers based on their importance during classification. Specifically, to measure the contribution score of each layer in predicting the target class, we employ the Layer-wise Relevance Propagation (LRP) method. LRP assigns local relevance at the output layer and propagates it through all layers, distributing the relevance until it reaches the input layers. These relevance scores serve as indicators for computing the layer contribution score. Additionally, we have introduced a clipped channel-wise quantization aimed at eliminating outliers from post-LayerNorm activations to alleviate severe inter-channel variations. To validate and assess our approach, we employ LRP-QViT across ViT, DeiT, and Swin transformer models on various datasets. Our experimental findings demonstrate that both our fixed-bit and mixed-bit post-training quantization methods surpass existing models in the context of 4-bit and 6-bit quantization.
Authors: Nikita Moriakov, Jan-Jakob Sonke, Jonas Teuwen
Cone Beam CT (CBCT) is an essential imaging modality nowadays, but the image quality of CBCT still lags behind the high quality standards established by the conventional Computed Tomography. We propose LIRE+, a learned iterative scheme for fast and memory-efficient CBCT reconstruction, which is a substantially faster and more parameter-efficient alternative to the recently proposed LIRE method. LIRE+ is a rotationally-equivariant multiscale learned invertible primal-dual iterative scheme for CBCT reconstruction. Memory usage is optimized by relying on simple reversible residual networks in primal/dual cells and patch-wise computations inside the cells during forward and backward passes, while increased inference speed is achieved by making the primal-dual scheme multiscale so that the reconstruction process starts at low resolution and with low resolution primal/dual latent vectors. A LIRE+ model was trained and validated on a set of 260 + 22 thorax CT scans and tested using a set of 142 thorax CT scans with additional evaluation with and without finetuning on an out-of-distribution set of 79 Head and Neck (HN) CT scans. Our method surpasses classical and deep learning baselines, including LIRE, on the thorax test set. For a similar inference time and with only 37 % of the parameter budget, LIRE+ achieves a +0.2 dB PSNR improvement over LIRE, while being able to match the performance of LIRE in 45 % less inference time and with 28 % of the parameter budget. Rotational equivariance ensures robustness of LIRE+ to patient orientation, while LIRE and other deep learning baselines suffer from substantial performance degradation when patient orientation is unusual. On the HN dataset in the absence of finetuning, LIRE+ is generally comparable to LIRE in performance apart from a few outlier cases, whereas after identical finetuning LIRE+ demonstates a +1.02 dB PSNR improvement over LIRE.
Authors: Weiguo Lu, Xuan Wu, Deng Ding, Jinqiao Duan, Jirong Zhuang, Gangnan Yuan
Diffusion models (DMs) are a type of generative model that has a huge impact on image synthesis and beyond. They achieve state-of-the-art generation results in various generative tasks. A great diversity of conditioning inputs, such as text or bounding boxes, are accessible to control the generation. In this work, we propose a conditioning mechanism utilizing Gaussian mixture models (GMMs) as feature conditioning to guide the denoising process. Based on set theory, we provide a comprehensive theoretical analysis that shows that conditional latent distribution based on features and classes is significantly different, so that conditional latent distribution on features produces fewer defect generations than conditioning on classes. Two diffusion models conditioned on the Gaussian mixture model are trained separately for comparison. Experiments support our findings. A novel gradient function called the negative Gaussian mixture gradient (NGMG) is proposed and applied in diffusion model training with an additional classifier. Training stability has improved. We also theoretically prove that NGMG shares the same benefit as the Earth Mover distance (Wasserstein) as a more sensible cost function when learning distributions supported by low-dimensional manifolds.
Authors: Mostafa Kazemi, Mahdi Rezaei, Mohsen Azarmi
As automated driving technology advances, the role of the driver to resume control of the vehicle in conditionally automated vehicles becomes increasingly critical. In the SAE Level 3 or partly automated vehicles, the driver needs to be available and ready to intervene when necessary. This makes it essential to evaluate their readiness accurately. This article presents a comprehensive analysis of driver readiness assessment by combining head pose features and eye-tracking data. The study explores the effectiveness of predictive models in evaluating driver readiness, addressing the challenges of dataset limitations and limited ground truth labels. Machine learning techniques, including LSTM architectures, are utilised to model driver readiness based on the Spatio-temporal status of the driver's head pose and eye gaze. The experiments in this article revealed that a Bidirectional LSTM architecture, combining both feature sets, achieves a mean absolute error of 0.363 on the DMD dataset, demonstrating superior performance in assessing driver readiness. The modular architecture of the proposed model also allows the integration of additional driver-specific features, such as steering wheel activity, enhancing its adaptability and real-world applicability.
Authors: Baoyun Peng, Min Liu, Zhaoning Zhang, Kai Xu, Dongsheng Li
Face recognition has made significant progress in recent years due to deep convolutional neural networks (CNN). In many face recognition (FR) scenarios, face images are acquired from a sequence with huge intra-variations. These intra-variations, which are mainly affected by the low-quality face images, cause instability of recognition performance. Previous works have focused on ad-hoc methods to select frames from a video or use face image quality assessment (FIQA) methods, which consider only a particular or combination of several distortions.
In this work, we present an efficient non-reference image quality assessment for FR that directly links image quality assessment (IQA) and FR. More specifically, we propose a new measurement to evaluate image quality without any reference. Based on the proposed quality measurement, we propose a deep Tiny Face Quality network (tinyFQnet) to learn a quality prediction function from data.
We evaluate the proposed method for different powerful FR models on two classical video-based (or template-based) benchmark: IJB-B and YTF. Extensive experiments show that, although the tinyFQnet is much smaller than the others, the proposed method outperforms state-of-the-art quality assessment methods in terms of effectiveness and efficiency.
Authors: Michael D. White, Alexander Tarakanov, Christopher P. Race, Philip J. Withers, Kody J.H. Law
Finding efficient means of fingerprinting microstructural information is a critical step towards harnessing data-centric machine learning approaches. A statistical framework is systematically developed for compressed characterisation of a population of images, which includes some classical computer vision methods as special cases. The focus is on materials microstructure. The ultimate purpose is to rapidly fingerprint sample images in the context of various high-throughput design/make/test scenarios. This includes, but is not limited to, quantification of the disparity between microstructures for quality control, classifying microstructures, predicting materials properties from image data and identifying potential processing routes to engineer new materials with specific properties. Here, we consider microstructure classification and utilise the resulting features over a range of related machine learning tasks, namely supervised, semi-supervised, and unsupervised learning.
The approach is applied to two distinct datasets to illustrate various aspects and some recommendations are made based on the findings. In particular, methods that leverage transfer learning with convolutional neural networks (CNNs), pretrained on the ImageNet dataset, are generally shown to outperform other methods. Additionally, dimensionality reduction of these CNN-based fingerprints is shown to have negligible impact on classification accuracy for the supervised learning approaches considered. In situations where there is a large dataset with only a handful of images labelled, graph-based label propagation to unlabelled data is shown to be favourable over discarding unlabelled data and performing supervised learning. In particular, label propagation by Poisson learning is shown to be highly effective at low label rates.
Authors: Weiguang Zhao, Chaolong Yang, Jianan Ye, Rui Zhang, Yuyao Yan, Xi Yang, Bin Dong, Amir Hussain, Kaizhu Huang
While weakly supervised multi-view face reconstruction (MVR) is garnering increased attention, one critical issue still remains open: how to effectively fuse multiple image information to reconstruct high-precision 3D models. In this regard, we propose a novel model called Deep Fusion MVR (DF-MVR) to reconstruct high-precision 3D facial shapes from multi-view images. Specifically, we introduce MulEn-Unet, a multi-view encoding to single decoding framework with skip connections and attention. This design allows for the extraction, integration, and compensation of deep features with attention from multi-view images. Furthermore, we adopt the involution kernel to enrich deep fusion features with channel features. In addition, we develop the face parse network to learn, identify, and emphasize the critical common face area within multi-view images. Experiments on Pixel-Face and Bosphorus datasets indicate the superiority of our model. Without 3D annotation, DF-MVR achieves 5.2% and 3.0% RMSE improvement over the existing weakly supervised MVRs respectively on Pixel-Face and Bosphorus dataset. Code will be available publicly at https://github.com/weiguangzhao/DF_MVR.
Authors: Feng Liang, Yangguang Li, Diana Marculescu
Recently, self-supervised Masked Autoencoders (MAE) have attracted unprecedented attention for their impressive representation learning ability. However, the pretext task, Masked Image Modeling (MIM), reconstructs the missing local patches, lacking the global understanding of the image. This paper extends MAE to a fully supervised setting by adding a supervised classification branch, thereby enabling MAE to learn global features from golden labels effectively. The proposed Supervised MAE (SupMAE) only exploits a visible subset of image patches for classification, unlike the standard supervised pre-training where all image patches are used. Through experiments, we demonstrate that SupMAE is not only more training efficient but it also learns more robust and transferable features. Specifically, SupMAE achieves comparable performance with MAE using only 30% of compute when evaluated on ImageNet with the ViT-B/16 model. SupMAE's robustness on ImageNet variants and transfer learning performance outperforms MAE and standard supervised pre-training counterparts. Codes are available at https://github.com/enyac-group/supmae.
Authors: Ruiming Du, Zhihong Ma, Pengyao Xie, Yong He, Haiyan Cen
Segmentation of plant point clouds to obtain high-precise morphological traits is essential for plant phenotyping. Although the fast development of deep learning has boosted much research on segmentation of plant point clouds, previous studies mainly focus on the hard voxelization-based or down-sampling-based methods, which are limited to segmenting simple plant organs. Segmentation of complex plant point clouds with a high spatial resolution still remains challenging. In this study, we proposed a deep learning network plant segmentation transformer (PST) to achieve the semantic and instance segmentation of rapeseed plants point clouds acquired by handheld laser scanning (HLS) with the high spatial resolution, which can characterize the tiny siliques as the main traits targeted. PST is composed of: (i) a dynamic voxel feature encoder (DVFE) to aggregate the point features with the raw spatial resolution; (ii) the dual window sets attention blocks to capture the contextual information; and (iii) a dense feature propagation module to obtain the final dense point feature map. The results proved that PST and PST-PointGroup (PG) achieved superior performance in semantic and instance segmentation tasks. For the semantic segmentation, the mean IoU, mean Precision, mean Recall, mean F1-score, and overall accuracy of PST were 93.96%, 97.29%, 96.52%, 96.88%, and 97.07%, achieving an improvement of 7.62%, 3.28%, 4.8%, 4.25%, and 3.88% compared to the second-best state-of-the-art network PAConv. For instance segmentation, PST-PG reached 89.51%, 89.85%, 88.83% and 82.53% in mCov, mWCov, mPerc90, and mRec90, achieving an improvement of 2.93%, 2.21%, 1.99%, and 5.9% compared to the original PG. This study proves that the deep-learning-based point cloud segmentation method has a great potential for resolving dense plant point clouds with complex morphological traits.
Authors: Zhijie Wang, Masanori Suganuma, Takayuki Okatani
Unsupervised domain adaptation (UDA) adapts a model trained on one domain (called source) to a novel domain (called target) using only unlabeled data. Due to its high annotation cost, researchers have developed many UDA methods for semantic segmentation, which assume no labeled sample is available in the target domain. We question the practicality of this assumption for two reasons. First, after training a model with a UDA method, we must somehow verify the model before deployment. Second, UDA methods have at least a few hyper-parameters that need to be determined. The surest solution to these is to evaluate the model using validation data, i.e., a certain amount of labeled target-domain samples. This question about the basic assumption of UDA leads us to rethink UDA from a data-centric point of view. Specifically, we assume we have access to a minimum level of labeled data. Then, we ask how much is necessary to find good hyper-parameters of existing UDA methods. We then consider what if we use the same data for supervised training of the same model, e.g., finetuning. We conducted experiments to answer these questions with popular scenarios, {GTA5, SYNTHIA}$\rightarrow$Cityscapes. We found that i) choosing good hyper-parameters needs only a few labeled images for some UDA methods whereas a lot more for others; and ii) simple finetuning works surprisingly well; it outperforms many UDA methods if only several dozens of labeled images are available.
Authors: Chentao Cao, Zhuo-Xu Cui, Yue Wang, Shaonan Liu, Taijin Chen, Hairong Zheng, Dong Liang, Yanjie Zhu
Diffusion models with continuous stochastic differential equations (SDEs) have shown superior performances in image generation. It can serve as a deep generative prior to solving the inverse problem in magnetic resonance (MR) reconstruction. However, low-frequency regions of $k$-space data are typically fully sampled in fast MR imaging, while existing diffusion models are performed throughout the entire image or $k$-space, inevitably introducing uncertainty in the reconstruction of low-frequency regions. Additionally, existing diffusion models often demand substantial iterations to converge, resulting in time-consuming reconstructions. To address these challenges, we propose a novel SDE tailored specifically for MR reconstruction with the diffusion process in high-frequency space (referred to as HFS-SDE). This approach ensures determinism in the fully sampled low-frequency regions and accelerates the sampling procedure of reverse diffusion. Experiments conducted on the publicly available fastMRI dataset demonstrate that the proposed HFS-SDE method outperforms traditional parallel imaging methods, supervised deep learning, and existing diffusion models in terms of reconstruction accuracy and stability. The fast convergence properties are also confirmed through theoretical and experimental validation. Our code and weights are available at https://github.com/Aboriginer/HFS-SDE.
Authors: Jonas Ricker, Simon Damm, Thorsten Holz, Asja Fischer
In the course of the past few years, diffusion models (DMs) have reached an unprecedented level of visual quality. However, relatively little attention has been paid to the detection of DM-generated images, which is critical to prevent adverse impacts on our society. In contrast, generative adversarial networks (GANs), have been extensively studied from a forensic perspective. In this work, we therefore take the natural next step to evaluate whether previous methods can be used to detect images generated by DMs. Our experiments yield two key findings: (1) state-of-the-art GAN detectors are unable to reliably distinguish real from DM-generated images, but (2) re-training them on DM-generated images allows for almost perfect detection, which remarkably even generalizes to GANs. Together with a feature space analysis, our results lead to the hypothesis that DMs produce fewer detectable artifacts and are thus more difficult to detect compared to GANs. One possible reason for this is the absence of grid-like frequency artifacts in DM-generated images, which are a known weakness of GANs. However, we make the interesting observation that diffusion models tend to underestimate high frequencies, which we attribute to the learning objective.
Authors: Gongping Chen, Lu Zhou, Jianxun Zhang, Xiaotao Yin, Liang Cui, Yu Dai
Breast cancer is one of the common cancers that endanger the health of women globally. Accurate target lesion segmentation is essential for early clinical intervention and postoperative follow-up. Recently, many convolutional neural networks (CNNs) have been proposed to segment breast tumors from ultrasound images. However, the complex ultrasound pattern and the variable tumor shape and size bring challenges to the accurate segmentation of the breast lesion. Motivated by the selective kernel convolution, we introduce an enhanced selective kernel convolution for breast tumor segmentation, which integrates multiple feature map region representations and adaptively recalibrates the weights of these feature map regions from the channel and spatial dimensions. This region recalibration strategy enables the network to focus more on high-contributing region features and mitigate the perturbation of less useful regions. Finally, the enhanced selective kernel convolution is integrated into U-net with deep supervision constraints to adaptively capture the robust representation of breast tumors. Extensive experiments with twelve state-of-the-art deep learning segmentation methods on three public breast ultrasound datasets demonstrate that our method has a more competitive segmentation performance in breast ultrasound images.
Authors: Yiyang Chen, Zhedong Zheng, Wei Ji, Leigang Qu, Tat-Seng Chua
We investigate composed image retrieval with text feedback. Users gradually look for the target of interest by moving from coarse to fine-grained feedback. However, existing methods merely focus on the latter, i.e., fine-grained search, by harnessing positive and negative pairs during training. This pair-based paradigm only considers the one-to-one distance between a pair of specific points, which is not aligned with the one-to-many coarse-grained retrieval process and compromises the recall rate. In an attempt to fill this gap, we introduce a unified learning approach to simultaneously modeling the coarse- and fine-grained retrieval by considering the multi-grained uncertainty. The key idea underpinning the proposed method is to integrate fine- and coarse-grained retrieval as matching data points with small and large fluctuations, respectively. Specifically, our method contains two modules: uncertainty modeling and uncertainty regularization. (1) The uncertainty modeling simulates the multi-grained queries by introducing identically distributed fluctuations in the feature space. (2) Based on the uncertainty modeling, we further introduce uncertainty regularization to adapt the matching objective according to the fluctuation range. Compared with existing methods, the proposed strategy explicitly prevents the model from pushing away potential candidates in the early stage, and thus improves the recall rate. On the three public datasets, i.e., FashionIQ, Fashion200k, and Shoes, the proposed method has achieved +4.03%, +3.38%, and +2.40% Recall@50 accuracy over a strong baseline, respectively.
Authors: Yu-Hsiang Wang, Jun-Wei Hsieh, Ping-Yang Chen, Ming-Ching Chang, Hung Hin So, Xin Li
Despite recent progress in Multiple Object Tracking (MOT), several obstacles such as occlusions, similar objects, and complex scenes remain an open challenge. Meanwhile, a systematic study of the cost-performance tradeoff for the popular tracking-by-detection paradigm is still lacking. This paper introduces SMILEtrack, an innovative object tracker that effectively addresses these challenges by integrating an efficient object detector with a Siamese network-based Similarity Learning Module (SLM). The technical contributions of SMILETrack are twofold. First, we propose an SLM that calculates the appearance similarity between two objects, overcoming the limitations of feature descriptors in Separate Detection and Embedding (SDE) models. The SLM incorporates a Patch Self-Attention (PSA) block inspired by the vision Transformer, which generates reliable features for accurate similarity matching. Second, we develop a Similarity Matching Cascade (SMC) module with a novel GATE function for robust object matching across consecutive video frames, further enhancing MOT performance. Together, these innovations help SMILETrack achieve an improved trade-off between the cost ({\em e.g.}, running speed) and performance (e.g., tracking accuracy) over several existing state-of-the-art benchmarks, including the popular BYTETrack method. SMILETrack outperforms BYTETrack by 0.4-0.8 MOTA and 2.1-2.2 HOTA points on MOT17 and MOT20 datasets. Code is available at https://github.com/pingyang1117/SMILEtrack_Official
Authors: Offry Hayon, Stefan Münger, Ilan Shimshoni, Ayellet Tal
Archaeology is an intriguing domain for computer vision. It suffers not only from shortage in (labeled) data, but also from highly-challenging data, which is often extremely abraded and damaged. This paper proposes a novel semi-supervised model for classification and retrieval of images of archaeological artifacts. This model utilizes unique data that exists in the domain -- manual drawings made by special artists. These are used during training to implicitly transfer the domain knowledge from the drawings to their corresponding images, improving their classification results. We show that while learning how to classify, our model also learns how to generate drawings of the artifacts, an important documentation task, which is currently performed manually. Last but not least, we collected a new dataset of stamp-seals of the Southern Levant. Our code and dataset are publicly available.
Authors: Zenglin Shi, Jing Jie, Ying Sun, Joo Hwee Lim, Mengmi Zhang
In AI, generalization refers to a model's ability to perform well on out-of-distribution data related to the given task, beyond the data it was trained on. For an AI agent to excel, it must also possess the continual learning capability, whereby an agent incrementally learns to perform a sequence of tasks without forgetting the previously acquired knowledge to solve the old tasks. Intuitively, generalization within a task allows the model to learn underlying features that can readily be applied to novel tasks, facilitating quicker learning and enhanced performance in subsequent tasks within a continual learning framework. Conversely, continual learning methods often include mechanisms to mitigate catastrophic forgetting, ensuring that knowledge from earlier tasks is retained. This preservation of knowledge over tasks plays a role in enhancing generalization for the ongoing task at hand. Despite the intuitive appeal of the interplay of both abilities, existing literature on continual learning and generalization has proceeded separately. In the preliminary effort to promote studies that bridge both fields, we first present empirical evidence showing that each of these fields has a mutually positive effect on the other. Next, building upon this finding, we introduce a simple and effective technique known as Shape-Texture Consistency Regularization (STCR), which caters to continual learning. STCR learns both shape and texture representations for each task, consequently enhancing generalization and thereby mitigating forgetting. Remarkably, extensive experiments validate that our STCR, can be seamlessly integrated with existing continual learning methods, where its performance surpasses these continual learning methods in isolation or when combined with established generalization techniques by a large margin. Our data and source code will be made publicly available upon publication.
Authors: Elad Hirsch, Ayellet Tal
Controllable image captioning models generate human-like image descriptions, enabling some kind of control over the generated captions. This paper focuses on controlling the caption length, i.e. a short and concise description or a long and detailed one. Since existing image captioning datasets contain mostly short captions, generating long captions is challenging. To address the shortage of long training examples, we propose to enrich the dataset with varying-length self-generated captions. These, however, might be of varying quality and are thus unsuitable for conventional training. We introduce a novel training strategy that selects the data points to be used at different times during the training. Our method dramatically improves the length-control abilities, while exhibiting SoTA performance in terms of caption quality. Our approach is general and is shown to be applicable also to paragraph generation.
Authors: Zhiwei Lin, Yongtao Wang, Shengxiang Qi, Nan Dong, Ming-Hsuan Yang
Existing LiDAR-based 3D object detection methods for autonomous driving scenarios mainly adopt the training-from-scratch paradigm. Unfortunately, this paradigm heavily relies on large-scale labeled data, whose collection can be expensive and time-consuming. Self-supervised pre-training is an effective and desirable way to alleviate this dependence on extensive annotated data. In this work, we present BEV-MAE, an efficient masked autoencoder pre-training framework for LiDAR-based 3D object detection in autonomous driving. Specifically, we propose a bird's eye view (BEV) guided masking strategy to guide the 3D encoder learning feature representation in a BEV perspective and avoid complex decoder design during pre-training. Furthermore, we introduce a learnable point token to maintain a consistent receptive field size of the 3D encoder with fine-tuning for masked point cloud inputs. Based on the property of outdoor point clouds in autonomous driving scenarios, i.e., the point clouds of distant objects are more sparse, we propose point density prediction to enable the 3D encoder to learn location information, which is essential for object detection. Experimental results show that BEV-MAE surpasses prior state-of-the-art self-supervised methods and achieves a favorably pre-training efficiency. Furthermore, based on TransFusion-L, BEV-MAE achieves new state-of-the-art LiDAR-based 3D object detection results, with 73.6 NDS and 69.6 mAP on the nuScenes benchmark. The source code will be released at https://github.com/VDIGPKU/BEV-MAE
Authors: Hao Feng, Keyi Zhou, Wengang Zhou, Yufei Yin, Jiajun Deng, Qi Sun, Houqiang Li
Contour-based instance segmentation has been actively studied, thanks to its flexibility and elegance in processing visual objects within complex backgrounds. In this work, we propose a novel deep network architecture, i.e., PolySnake, for generic contour-based instance segmentation. Motivated by the classic Snake algorithm, the proposed PolySnake achieves superior and robust segmentation performance with an iterative and progressive contour refinement strategy. Technically, PolySnake introduces a recurrent update operator to estimate the object contour iteratively. It maintains a single estimate of the contour that is progressively deformed toward the object boundary. At each iteration, PolySnake builds a semantic-rich representation for the current contour and feeds it to the recurrent operator for further contour adjustment. Through the iterative refinements, the contour progressively converges to a stable status that tightly encloses the object instance. Beyond the scope of general instance segmentation, extensive experiments are conducted to validate the effectiveness and generalizability of our PolySnake in two additional specific task scenarios, including scene text detection and lane detection. The results demonstrate that the proposed PolySnake outperforms the existing advanced methods on several multiple prevalent benchmarks across the three tasks. The codes and pre-trained models are available at https://github.com/fh2019ustc/PolySnake
Authors: Minh-Quan Le, Tam V. Nguyen, Trung-Nghia Le, Thanh-Toan Do, Minh N. Do, Minh-Triet Tran
Few-shot instance segmentation extends the few-shot learning paradigm to the instance segmentation task, which tries to segment instance objects from a query image with a few annotated examples of novel categories. Conventional approaches have attempted to address the task via prototype learning, known as point estimation. However, this mechanism depends on prototypes (\eg mean of $K-$shot) for prediction, leading to performance instability. To overcome the disadvantage of the point estimation mechanism, we propose a novel approach, dubbed MaskDiff, which models the underlying conditional distribution of a binary mask, which is conditioned on an object region and $K-$shot information. Inspired by augmentation approaches that perturb data with Gaussian noise for populating low data density regions, we model the mask distribution with a diffusion probabilistic model. We also propose to utilize classifier-free guided mask sampling to integrate category information into the binary mask generation process. Without bells and whistles, our proposed method consistently outperforms state-of-the-art methods on both base and novel classes of the COCO dataset while simultaneously being more stable than existing methods. The source code is available at: https://github.com/minhquanlecs/MaskDiff.
Authors: Anastasiia Kornilova, Ivan Moskalenko, Timofei Pushkin, Fakhriddin Tojiboev, Rahim Tariverdizadeh, Gonzalo Ferrer
This paper presents an approach for creating a visual place recognition (VPR) database for localization in indoor environments from RGBD scanning sequences. The proposed approach is formulated as a minimization problem in terms of dominating set algorithm for graph, constructed from spatial information, and referred as DominatingSet. Our algorithm shows better scene coverage in comparison to other methodologies that are used for database creation. Also, we demonstrate that using DominatingSet, a database size could be up to 250-1400 times smaller than the original scanning sequence while maintaining a recall rate of more than 80% on testing sequences. We evaluated our algorithm on 7-scenes and BundleFusion datasets and an additionally recorded sequence in a highly repetitive office setting. In addition, the database selection can produce weakly-supervised labels for fine-tuning neural place recognition algorithms to particular settings, improving even more their accuracy. The paper also presents a fully automated pipeline for VPR database creation from RGBD scanning sequences, as well as a set of metrics for VPR database evaluation. The code and released data are available on our web-page~ -- https://prime-slam.github.io/place-recognition-db/
Authors: Leichao Cui, Xiuxian Li, Min Meng, Xiaoyu Mo
LiDAR point clouds have become the most common data source in autonomous driving. However, due to the sparsity of point clouds, accurate and reliable detection cannot be achieved in specific scenarios. Because of their complementarity with point clouds, images are getting increasing attention. Although with some success, existing fusion methods either perform hard fusion or do not fuse in a direct manner. In this paper, we propose a generic 3D detection framework called MMFusion, using multi-modal features. The framework aims to achieve accurate fusion between LiDAR and images to improve 3D detection in complex scenes. Our framework consists of two separate streams: the LiDAR stream and the camera stream, which can be compatible with any single-modal feature extraction network. The Voxel Local Perception Module in the LiDAR stream enhances local feature representation, and then the Multi-modal Feature Fusion Module selectively combines feature output from different streams to achieve better fusion. Extensive experiments have shown that our framework not only outperforms existing benchmarks but also improves their detection, especially for detecting cyclists and pedestrians on KITTI benchmarks, with strong robustness and generalization capabilities. Hopefully, our work will stimulate more research into multi-modal fusion for autonomous driving tasks.
Authors: Weijia Wu, Yuzhong Zhao, Mike Zheng Shou, Hong Zhou, Chunhua Shen
Collecting and annotating images with pixel-wise labels is time-consuming and laborious. In contrast, synthetic data can be freely available using a generative model (e.g., DALL-E, Stable Diffusion). In this paper, we show that it is possible to automatically obtain accurate semantic masks of synthetic images generated by the Off-the-shelf Stable Diffusion model, which uses only text-image pairs during training. Our approach, called DiffuMask, exploits the potential of the cross-attention map between text and image, which is natural and seamless to extend the text-driven image synthesis to semantic mask generation. DiffuMask uses text-guided cross-attention information to localize class/word-specific regions, which are combined with practical techniques to create a novel high-resolution and class-discriminative pixel-wise mask. The methods help to reduce data collection and annotation costs obviously. Experiments demonstrate that the existing segmentation methods trained on synthetic data of DiffuMask can achieve a competitive performance over the counterpart of real data (VOC 2012, Cityscapes). For some classes (e.g., bird), DiffuMask presents promising performance, close to the stateof-the-art result of real data (within 3% mIoU gap). Moreover, in the open-vocabulary segmentation (zero-shot) setting, DiffuMask achieves a new SOTA result on Unseen class of VOC 2012. The project website can be found at https://weijiawu.github.io/DiffusionMask/.
Authors: Willi Menapace, Aliaksandr Siarohin, Stéphane Lathuilière, Panos Achlioptas, Vladislav Golyanik, Sergey Tulyakov, Elisa Ricci
Neural video game simulators emerged as powerful tools to generate and edit videos. Their idea is to represent games as the evolution of an environment's state driven by the actions of its agents. While such a paradigm enables users to play a game action-by-action, its rigidity precludes more semantic forms of control. To overcome this limitation, we augment game models with prompts specified as a set of natural language actions and desired states. The result-a Promptable Game Model (PGM)-makes it possible for a user to play the game by prompting it with high- and low-level action sequences. Most captivatingly, our PGM unlocks the director's mode, where the game is played by specifying goals for the agents in the form of a prompt. This requires learning "game AI", encapsulated by our animation model, to navigate the scene using high-level constraints, play against an adversary, and devise a strategy to win a point. To render the resulting state, we use a compositional NeRF representation encapsulated in our synthesis model. To foster future research, we present newly collected, annotated and calibrated Tennis and Minecraft datasets. Our method significantly outperforms existing neural video game simulators in terms of rendering quality and unlocks applications beyond the capabilities of the current state of the art. Our framework, data, and models are available at https://snap-research.github.io/promptable-game-models/.
Authors: Dong An, Hanqing Wang, Wenguan Wang, Zun Wang, Yan Huang, Keji He, Liang Wang
Vision-language navigation is a task that requires an agent to follow instructions to navigate in environments. It becomes increasingly crucial in the field of embodied AI, with potential applications in autonomous navigation, search and rescue, and human-robot interaction. In this paper, we propose to address a more practical yet challenging counterpart setting - vision-language navigation in continuous environments (VLN-CE). To develop a robust VLN-CE agent, we propose a new navigation framework, ETPNav, which focuses on two critical skills: 1) the capability to abstract environments and generate long-range navigation plans, and 2) the ability of obstacle-avoiding control in continuous environments. ETPNav performs online topological mapping of environments by self-organizing predicted waypoints along a traversed path, without prior environmental experience. It privileges the agent to break down the navigation procedure into high-level planning and low-level control. Concurrently, ETPNav utilizes a transformer-based cross-modal planner to generate navigation plans based on topological maps and instructions. The plan is then performed through an obstacle-avoiding controller that leverages a trial-and-error heuristic to prevent navigation from getting stuck in obstacles. Experimental results demonstrate the effectiveness of the proposed method. ETPNav yields more than 10% and 20% improvements over prior state-of-the-art on R2R-CE and RxR-CE datasets, respectively. Our code is available at https://github.com/MarSaKi/ETPNav.
Authors: Mahdi S. Hosseini, Babak Ehteshami Bejnordi, Vincent Quoc-Huy Trinh, Danial Hasan, Xingwen Li, Taehyo Kim, Haochen Zhang, Theodore Wu, Kajanan Chinniah, Sina Maghsoudlou, Ryan Zhang, Stephen Yang, Jiadai Zhu, Lyndon Chan, Samir Khaki, Andrei Buin, Fatemeh Chaji, Ala Salehi, Bich Ngoc Nguyen, Dimitris Samaras, Konstantinos N. Plataniotis
Computational Pathology CPath is an interdisciplinary science that augments developments of computational approaches to analyze and model medical histopathology images. The main objective for CPath is to develop infrastructure and workflows of digital diagnostics as an assistive CAD system for clinical pathology, facilitating transformational changes in the diagnosis and treatment of cancer that are mainly address by CPath tools. With evergrowing developments in deep learning and computer vision algorithms, and the ease of the data flow from digital pathology, currently CPath is witnessing a paradigm shift. Despite the sheer volume of engineering and scientific works being introduced for cancer image analysis, there is still a considerable gap of adopting and integrating these algorithms in clinical practice. This raises a significant question regarding the direction and trends that are undertaken in CPath. In this article we provide a comprehensive review of more than 800 papers to address the challenges faced in problem design all-the-way to the application and implementation viewpoints. We have catalogued each paper into a model-card by examining the key works and challenges faced to layout the current landscape in CPath. We hope this helps the community to locate relevant works and facilitate understanding of the field's future directions. In a nutshell, we oversee the CPath developments in cycle of stages which are required to be cohesively linked together to address the challenges associated with such multidisciplinary science. We overview this cycle from different perspectives of data-centric, model-centric, and application-centric problems. We finally sketch remaining challenges and provide directions for future technical developments and clinical integration of CPath (https://github.com/AtlasAnalyticsLab/CPath_Survey).
Authors: Luyang Luo, Xi Wang, Yi Lin, Xiaoqi Ma, Andong Tan, Ronald Chan, Varut Vardhanabhuti, Winnie CW Chu, Kwang-Ting Cheng, Hao Chen
Breast cancer has reached the highest incidence rate worldwide among all malignancies since 2020. Breast imaging plays a significant role in early diagnosis and intervention to improve the outcome of breast cancer patients. In the past decade, deep learning has shown remarkable progress in breast cancer imaging analysis, holding great promise in interpreting the rich information and complex context of breast imaging modalities. Considering the rapid improvement in deep learning technology and the increasing severity of breast cancer, it is critical to summarize past progress and identify future challenges to be addressed. This paper provides an extensive review of deep learning-based breast cancer imaging research, covering studies on mammogram, ultrasound, magnetic resonance imaging, and digital pathology images over the past decade. The major deep learning methods and applications on imaging-based screening, diagnosis, treatment response prediction, and prognosis are elaborated and discussed. Drawn from the findings of this survey, we present a comprehensive discussion of the challenges and potential avenues for future research in deep learning-based breast cancer imaging.
Authors: Thanh-Danh Nguyen, Anh-Khoa Nguyen Vu, Nhat-Duy Nguyen, Vinh-Tiep Nguyen, Thanh Duc Ngo, Thanh-Toan Do, Minh-Triet Tran, Tam V. Nguyen
Camouflaged object detection and segmentation is a new and challenging research topic in computer vision. There is a serious issue of lacking data of camouflaged objects such as camouflaged animals in natural scenes. In this paper, we address the problem of few-shot learning for camouflaged object detection and segmentation. To this end, we first collect a new dataset, CAMO-FS, for the benchmark. We then propose a novel method to efficiently detect and segment the camouflaged objects in the images. In particular, we introduce the instance triplet loss and the instance memory storage. The extensive experiments demonstrated that our proposed method achieves state-of-the-art performance on the newly collected dataset.
Authors: Junge Zhang, Feihu Zhang, Shaochen Kuang, Li Zhang
Labeling LiDAR point clouds for training autonomous driving is extremely expensive and difficult. LiDAR simulation aims at generating realistic LiDAR data with labels for training and verifying self-driving algorithms more efficiently. Recently, Neural Radiance Fields (NeRF) have been proposed for novel view synthesis using implicit reconstruction of 3D scenes. Inspired by this, we present NeRF-LIDAR, a novel LiDAR simulation method that leverages real-world information to generate realistic LIDAR point clouds. Different from existing LiDAR simulators, we use real images and point cloud data collected by self-driving cars to learn the 3D scene representation, point cloud generation and label rendering. We verify the effectiveness of our NeRF-LiDAR by training different 3D segmentation models on the generated LiDAR point clouds. It reveals that the trained models are able to achieve similar accuracy when compared with the same model trained on the real LiDAR data. Besides, the generated data is capable of boosting the accuracy through pre-training which helps reduce the requirements of the real labeled data.
Authors: Shanchuan Lin, Bingchen Liu, Jiashi Li, Xiao Yang
We discover that common diffusion noise schedules do not enforce the last timestep to have zero signal-to-noise ratio (SNR), and some implementations of diffusion samplers do not start from the last timestep. Such designs are flawed and do not reflect the fact that the model is given pure Gaussian noise at inference, creating a discrepancy between training and inference. We show that the flawed design causes real problems in existing implementations. In Stable Diffusion, it severely limits the model to only generate images with medium brightness and prevents it from generating very bright and dark samples. We propose a few simple fixes: (1) rescale the noise schedule to enforce zero terminal SNR; (2) train the model with v prediction; (3) change the sampler to always start from the last timestep; (4) rescale classifier-free guidance to prevent over-exposure. These simple changes ensure the diffusion process is congruent between training and inference and allow the model to generate samples more faithful to the original data distribution.
Authors: Omer Belhasin, Yaniv Romano, Daniel Freedman, Ehud Rivlin, Michael Elad
Uncertainty quantification for inverse problems in imaging has drawn much attention lately. Existing approaches towards this task define uncertainty regions based on probable values per pixel, while ignoring spatial correlations within the image, resulting in an exaggerated volume of uncertainty. In this paper, we propose PUQ (Principal Uncertainty Quantification) -- a novel definition and corresponding analysis of uncertainty regions that takes into account spatial relationships within the image, thus providing reduced volume regions. Using recent advancements in generative models, we derive uncertainty intervals around principal components of the empirical posterior distribution, forming an ambiguity region that guarantees the inclusion of true unseen values with a user-defined confidence probability. To improve computational efficiency and interpretability, we also guarantee the recovery of true unseen values using only a few principal directions, resulting in more informative uncertainty regions. Our approach is verified through experiments on image colorization, super-resolution, and inpainting; its effectiveness is shown through comparison to baseline methods, demonstrating significantly tighter uncertainty regions.
Authors: Xi Weng, Yunhao Ni, Tengwei Song, Jie Luo, Rao Muhammad Anwer, Salman Khan, Fahad Shahbaz Khan, Lei Huang
Whitening loss offers a theoretical guarantee against feature collapse in self-supervised learning (SSL) with joint embedding architectures. Typically, it involves a hard whitening approach, transforming the embedding and applying loss to the whitened output. In this work, we introduce Spectral Transformation (ST), a framework to modulate the spectrum of embedding and to seek for functions beyond whitening that can avoid dimensional collapse. We show that whitening is a special instance of ST by definition, and our empirical investigations unveil other ST instances capable of preventing collapse. Additionally, we propose a novel ST instance named IterNorm with trace loss (INTL). Theoretical analysis confirms INTL's efficacy in preventing collapse and modulating the spectrum of embedding toward equal-eigenvalues during optimization. Our experiments on ImageNet classification and COCO object detection demonstrate INTL's potential in learning superior representations. The code is available at https://github.com/winci-ai/INTL.
Authors: Sanyam Jain
This paper implements and investigates popular adversarial attacks on the YOLOv5 Object Detection algorithm. The paper explores the vulnerability of the YOLOv5 to adversarial attacks in the context of traffic and road sign detection. The paper investigates the impact of different types of attacks, including the Limited memory Broyden Fletcher Goldfarb Shanno (L-BFGS), the Fast Gradient Sign Method (FGSM) attack, the Carlini and Wagner (C&W) attack, the Basic Iterative Method (BIM) attack, the Projected Gradient Descent (PGD) attack, One Pixel Attack, and the Universal Adversarial Perturbations attack on the accuracy of YOLOv5 in detecting traffic and road signs. The results show that YOLOv5 is susceptible to these attacks, with misclassification rates increasing as the magnitude of the perturbations increases. We also explain the results using saliency maps. The findings of this paper have important implications for the safety and reliability of object detection algorithms used in traffic and transportation systems, highlighting the need for more robust and secure models to ensure their effectiveness in real-world applications.
Authors: Zhizhong Chai, Luyang Luo, Huangjing Lin, Pheng-Ann Heng, Hao Chen
Deep learning (DL)-based rib fracture detection has shown promise of playing an important role in preventing mortality and improving patient outcome. Normally, developing DL-based object detection models requires a huge amount of bounding box annotation. However, annotating medical data is time-consuming and expertise-demanding, making obtaining a large amount of fine-grained annotations extremely infeasible. This poses a pressing need {for} developing label-efficient detection models to alleviate radiologists' labeling burden. To tackle this challenge, the literature on object detection has witnessed an increase of weakly-supervised and semi-supervised approaches, yet still lacks a unified framework that leverages various forms of fully-labeled, weakly-labeled, and unlabeled data. In this paper, we present a novel omni-supervised object detection network, ORF-Netv2, to leverage as much available supervision as possible. Specifically, a multi-branch omni-supervised detection head is introduced with each branch trained with a specific type of supervision. A co-training-based dynamic label assignment strategy is then proposed to enable flexible and robust learning from the weakly-labeled and unlabeled data. Extensive evaluation was conducted for the proposed framework with three rib fracture datasets on both chest CT and X-ray. By leveraging all forms of supervision, ORF-Netv2 achieves mAPs of 34.7, 44.7, and 19.4 on the three datasets, respectively, surpassing the baseline detector which uses only box annotations by mAP gains of 3.8, 4.8, and 5.0, respectively. Furthermore, ORF-Netv2 consistently outperforms other competitive label-efficient methods over various scenarios, showing a promising framework for label-efficient fracture detection. The code is available at: https://github.com/zhizhongchai/ORF-Net.
Authors: Apratim Bhattacharyya, Sunny Panchal, Mingu Lee, Reza Pourreza, Pulkit Madan, Roland Memisevic
Multi-modal language models (LM) have recently shown promising performance in high-level reasoning tasks on videos. However, existing methods still fall short in tasks like causal or compositional spatiotemporal reasoning over actions, in which model predictions need to be grounded in fine-grained low-level details, such as object motions and object interactions. In this work, we propose training an LM end-to-end on low-level surrogate tasks, including object detection, re-identification, and tracking, to endow the model with the required low-level visual capabilities. We show that a two-stream video encoder with spatiotemporal attention is effective at capturing the required static and motion-based cues in the video. By leveraging the LM's ability to perform the low-level surrogate tasks, we can cast reasoning in videos as the three-step process of Look, Remember, Reason wherein visual information is extracted using low-level visual skills step-by-step and then integrated to arrive at a final answer. We demonstrate the effectiveness of our framework on diverse visual reasoning tasks from the ACRE, CATER, Something-Else and STAR datasets. Our approach is trainable end-to-end and surpasses state-of-the-art task-specific methods across these tasks by a large margin.
Authors: Ahmed Ghorbel, Wassim Hamidouche, Luce Morin
Recently, the performance of neural image compression (NIC) has steadily improved thanks to the last line of study, reaching or outperforming state-of-the-art conventional codecs. Despite significant progress, current NIC methods still rely on ConvNet-based entropy coding, limited in modeling long-range dependencies due to their local connectivity and the increasing number of architectural biases and priors, resulting in complex underperforming models with high decoding latency. Motivated by the efficiency investigation of the Tranformer-based transform coding framework, namely SwinT-ChARM, we propose to enhance the latter, as first, with a more straightforward yet effective Tranformer-based channel-wise auto-regressive prior model, resulting in an absolute image compression transformer (ICT). Through the proposed ICT, we can capture both global and local contexts from the latent representations and better parameterize the distribution of the quantized latents. Further, we leverage a learnable scaling module with a sandwich ConvNeXt-based pre-/post-processor to accurately extract more compact latent codes while reconstructing higher-quality images. Extensive experimental results on benchmark datasets showed that the proposed framework significantly improves the trade-off between coding efficiency and decoder complexity over the versatile video coding (VVC) reference encoder (VTM-18.0) and the neural codec SwinT-ChARM. Moreover, we provide model scaling studies to verify the computational efficiency of our approach and conduct several objective and subjective analyses to bring to the fore the performance gap between the adaptive image compression transformer (AICT) and the neural codec SwinT-ChARM.
Authors: Wen Yan, Bernard Chiu, Ziyi Shen, Qianye Yang, Tom Syer, Zhe Min, Shonit Punwani, Mark Emberton, David Atkinson, Dean C. Barratt, Yipeng Hu
One of the distinct characteristics in radiologists' reading of multiparametric prostate MR scans, using reporting systems such as PI-RADS v2.1, is to score individual types of MR modalities, T2-weighted, diffusion-weighted, and dynamic contrast-enhanced, and then combine these image-modality-specific scores using standardised decision rules to predict the likelihood of clinically significant cancer. This work aims to demonstrate that it is feasible for low-dimensional parametric models to model such decision rules in the proposed Combiner networks, without compromising the accuracy of predicting radiologic labels: First, it is shown that either a linear mixture model or a nonlinear stacking model is sufficient to model PI-RADS decision rules for localising prostate cancer. Second, parameters of these (generalised) linear models are proposed as hyperparameters, to weigh multiple networks that independently represent individual image modalities in the Combiner network training, as opposed to end-to-end modality ensemble. A HyperCombiner network is developed to train a single image segmentation network that can be conditioned on these hyperparameters during inference, for much improved efficiency. Experimental results based on data from 850 patients, for the application of automating radiologist labelling multi-parametric MR, compare the proposed combiner networks with other commonly-adopted end-to-end networks. Using the added advantages of obtaining and interpreting the modality combining rules, in terms of the linear weights or odds-ratios on individual image modalities, three clinical applications are presented for prostate cancer segmentation, including modality availability assessment, importance quantification and rule discovery.
Authors: Isabel Funke, Dominik Rivoir, Stefanie Krell, Stefanie Speidel
To enable context-aware computer assistance in the operating room of the future, cognitive systems need to understand automatically which surgical phase is being performed by the medical team. The primary source of information for surgical phase recognition is typically video, which presents two challenges: extracting meaningful features from the video stream and effectively modeling temporal information in the sequence of visual features. For temporal modeling, attention mechanisms have gained popularity due to their ability to capture long-range dependencies. In this paper, we explore design choices for attention in existing temporal models for surgical phase recognition and propose a novel approach that uses attention more effectively: TUNeS, an efficient and simple temporal model that incorporates self-attention at the core of a convolutional U-Net structure. In addition, we propose to train the feature extractor, a standard CNN, together with an LSTM on preferably long video segments, i.e., with long temporal context. In our experiments, all temporal models performed better on top of feature extractors that were trained with longer temporal context. On these contextualized features, TUNeS achieves state-of-the-art results on the Cholec80 dataset.
Authors: Li Li, Wei Ji, Yiming Wu, Mengze Li, You Qin, Lina Wei, Roger Zimmermann
Panoptic Scene Graph Generation (PSG) parses objects and predicts their relationships (predicate) to connect human language and visual scenes. However, different language preferences of annotators and semantic overlaps between predicates lead to biased predicate annotations in the dataset, i.e. different predicates for same object pairs. Biased predicate annotations make PSG models struggle in constructing a clear decision plane among predicates, which greatly hinders the real application of PSG models. To address the intrinsic bias above, we propose a novel framework named ADTrans to adaptively transfer biased predicate annotations to informative and unified ones. To promise consistency and accuracy during the transfer process, we propose to measure the invariance of representations in each predicate class, and learn unbiased prototypes of predicates with different intensities. Meanwhile, we continuously measure the distribution changes between each presentation and its prototype, and constantly screen potential biased data. Finally, with the unbiased predicate-prototype representation embedding space, biased annotations are easily identified. Experiments show that ADTrans significantly improves the performance of benchmark models, achieving a new state-of-the-art performance, and shows great generalization and effectiveness on multiple datasets.
Authors: Mingzhan Yang, Guangxin Han, Bin Yan, Wenhua Zhang, Jinqing Qi, Huchuan Lu, Dong Wang
Multi-Object Tracking (MOT) aims to detect and associate all desired objects across frames. Most methods accomplish the task by explicitly or implicitly leveraging strong cues (i.e., spatial and appearance information), which exhibit powerful instance-level discrimination. However, when object occlusion and clustering occur, spatial and appearance information will become ambiguous simultaneously due to the high overlap among objects. In this paper, we demonstrate this long-standing challenge in MOT can be efficiently and effectively resolved by incorporating weak cues to compensate for strong cues. Along with velocity direction, we introduce the confidence and height state as potential weak cues. With superior performance, our method still maintains Simple, Online and Real-Time (SORT) characteristics. Also, our method shows strong generalization for diverse trackers and scenarios in a plug-and-play and training-free manner. Significant and consistent improvements are observed when applying our method to 5 different representative trackers. Further, with both strong and weak cues, our method Hybrid-SORT achieves superior performance on diverse benchmarks, including MOT17, MOT20, and especially DanceTrack where interaction and severe occlusion frequently happen with complex motions. The code and models are available at https://github.com/ymzis69/HybridSORT.
Authors: Yeying Jin, Beibei Lin, Wending Yan, Yuan Yuan, Wei Ye, Robby T. Tan
Visibility in hazy nighttime scenes is frequently reduced by multiple factors, including low light, intense glow, light scattering, and the presence of multicolored light sources. Existing nighttime dehazing methods often struggle with handling glow or low-light conditions, resulting in either excessively dark visuals or unsuppressed glow outputs. In this paper, we enhance the visibility from a single nighttime haze image by suppressing glow and enhancing low-light regions. To handle glow effects, our framework learns from the rendered glow pairs. Specifically, a light source aware network is proposed to detect light sources of night images, followed by the APSF (Atmospheric Point Spread Function)-guided glow rendering. Our framework is then trained on the rendered images, resulting in glow suppression. Moreover, we utilize gradient-adaptive convolution, to capture edges and textures in hazy scenes. By leveraging extracted edges and textures, we enhance the contrast of the scene without losing important structural details. To boost low-light intensity, our network learns an attention map, then adjusted by gamma correction. This attention has high values on low-light regions and low values on haze and glow regions. Extensive evaluation on real nighttime haze images, demonstrates the effectiveness of our method. Our experiments demonstrate that our method achieves a PSNR of 30.38dB, outperforming state-of-the-art methods by 13% on GTA5 nighttime haze dataset. Our data and code is available at https://github.com/jinyeying/nighttime_dehaze.
Authors: Yu-Ju Tsai, Yu-Lun Liu, Lu Qi, Kelvin C.K. Chan, Ming-Hsuan Yang
Restoring facial details from low-quality (LQ) images has remained a challenging problem due to its ill-posedness induced by various degradations in the wild. The existing codebook prior mitigates the ill-posedness by leveraging an autoencoder and learned codebook of high-quality (HQ) features, achieving remarkable quality. However, existing approaches in this paradigm frequently depend on a single encoder pre-trained on HQ data for restoring HQ images, disregarding the domain gap between LQ and HQ images. As a result, the encoding of LQ inputs may be insufficient, resulting in suboptimal performance. To tackle this problem, we propose a novel dual-branch framework named DAEFR. Our method introduces an auxiliary LQ branch that extracts crucial information from the LQ inputs. Additionally, we incorporate association training to promote effective synergy between the two branches, enhancing code prediction and output quality. We evaluate the effectiveness of DAEFR on both synthetic and real-world datasets, demonstrating its superior performance in restoring facial details. Project page: https://liagm.github.io/DAEFR/
Authors: Yubiao Yue, Xinyu Zeng, Xiaoqiang Shi, Meiping Zhang, Haihua Liang, Fan Zhang, Yanmei Chen, Zefeng Xie, Wenrui Wu, Zhenzhang Li
Traditional ear disease diagnosis heavily depends on experienced specialists and specialized equipment, frequently resulting in misdiagnoses, treatment delays, and financial burdens for some patients. Utilizing deep learning models for efficient ear disease diagnosis has proven effective and affordable. However, existing research overlooked model inference speed and parameter size required for deployment. To tackle these challenges, we constructed a large-scale dataset comprising eight ear disease categories and normal ear canal samples from two hospitals. Inspired by ShuffleNetV2, we developed Best-EarNet, an ultrafast and ultralight network enabling real-time ear disease diagnosis. Best-EarNet incorporates the novel Local-Global Spatial Feature Fusion Module which can capture global and local spatial information simultaneously and guide the network to focus on crucial regions within feature maps at various levels, mitigating low accuracy issues. Moreover, our network uses multiple auxiliary classification heads for efficient parameter optimization. With 0.77M parameters, Best-EarNet achieves an average frames per second of 80 on CPU. Employing transfer learning and five-fold cross-validation with 22,581 images from Hospital-1, the model achieves an impressive 95.23% accuracy. External testing on 1,652 images from Hospital-2 validates its performance, yielding 92.14% accuracy. Compared to state-of-the-art networks, Best-EarNet establishes a new state-of-the-art (SOTA) in practical applications. Most importantly, we developed an intelligent diagnosis system called Ear Keeper, which can be deployed on common electronic devices. By manipulating a compact electronic otoscope, users can perform comprehensive scanning and diagnosis of the ear canal using real-time video. This study provides a novel paradigm for ear endoscopy and other medical endoscopic image recognition applications.
Authors: Jiacong Xu, Yi Zhang, Jiawei Peng, Wufei Ma, Artur Jesslen, Pengliang Ji, Qixin Hu, Jiehua Zhang, Qihao Liu, Jiahao Wang, Wei Ji, Chen Wang, Xiaoding Yuan, Prakhar Kaushik, Guofeng Zhang, Jie Liu, Yushan Xie, Yawen Cui, Alan Yuille, Adam Kortylewski
Accurately estimating the 3D pose and shape is an essential step towards understanding animal behavior, and can potentially benefit many downstream applications, such as wildlife conservation. However, research in this area is held back by the lack of a comprehensive and diverse dataset with high-quality 3D pose and shape annotations. In this paper, we propose Animal3D, the first comprehensive dataset for mammal animal 3D pose and shape estimation. Animal3D consists of 3379 images collected from 40 mammal species, high-quality annotations of 26 keypoints, and importantly the pose and shape parameters of the SMAL model. All annotations were labeled and checked manually in a multi-stage process to ensure highest quality results. Based on the Animal3D dataset, we benchmark representative shape and pose estimation models at: (1) supervised learning from only the Animal3D data, (2) synthetic to real transfer from synthetically generated images, and (3) fine-tuning human pose and shape estimation models. Our experimental results demonstrate that predicting the 3D shape and pose of animals across species remains a very challenging task, despite significant advances in human pose estimation. Our results further demonstrate that synthetic pre-training is a viable strategy to boost the model performance. Overall, Animal3D opens new directions for facilitating future research in animal 3D pose and shape estimation, and is publicly available.
Authors: Saeid Asgari Taghanaki, Aliasghar Khani, Amir Khasahmadi, Aditya Sanghi, Karl D.D. Willis, Ali Mahdavi-Amiri
Interpreting the learned features of vision models has posed a longstanding challenge in the field of machine learning. To address this issue, we propose a novel method that leverages the capabilities of language models to interpret the learned features of pre-trained image classifiers. Our method, called TExplain, tackles this task by training a neural network to establish a connection between the feature space of image classifiers and language models. Then, during inference, our approach generates a vast number of sentences to explain the features learned by the classifier for a given image. These sentences are then used to extract the most frequent words, providing a comprehensive understanding of the learned features and patterns within the classifier. Our method, for the first time, utilizes these frequent words corresponding to a visual representation to provide insights into the decision-making process of the independently trained classifier, enabling the detection of spurious correlations, biases, and a deeper comprehension of its behavior. To validate the effectiveness of our approach, we conduct experiments on diverse datasets, including ImageNet-9L and Waterbirds. The results demonstrate the potential of our method to enhance the interpretability and robustness of image classifiers.
Authors: Minsu Kim, Jaewon Lee, Byeonghun Lee, Sunghoon Im, Kyong Hwan Jin
Existing frameworks for image stitching often provide visually reasonable stitchings. However, they suffer from blurry artifacts and disparities in illumination, depth level, etc. Although the recent learning-based stitchings relax such disparities, the required methods impose sacrifice of image qualities failing to capture high-frequency details for stitched images. To address the problem, we propose a novel approach, implicit Neural Image Stitching (NIS) that extends arbitrary-scale super-resolution. Our method estimates Fourier coefficients of images for quality-enhancing warps. Then, the suggested model blends color mismatches and misalignment in the latent space and decodes the features into RGB values of stitched images. Our experiments show that our approach achieves improvement in resolving the low-definition imaging of the previous deep image stitching with favorable accelerated image-enhancing methods. Our source code is available at https://github.com/minshu-kim/NIS.
Authors: Jinglong Wang, Xiawei Li, Jing Zhang, Qingyuan Xu, Qin Zhou, Qian Yu, Lu Sheng, Dong Xu
The pre-trained text-image discriminative models, such as CLIP, has been explored for open-vocabulary semantic segmentation with unsatisfactory results due to the loss of crucial localization information and awareness of object shapes. Recently, there has been a growing interest in expanding the application of generative models from generation tasks to semantic segmentation. These approaches utilize generative models either for generating annotated data or extracting features to facilitate semantic segmentation. This typically involves generating a considerable amount of synthetic data or requiring additional mask annotations. To this end, we uncover the potential of generative text-to-image diffusion models (e.g., Stable Diffusion) as highly efficient open-vocabulary semantic segmenters, and introduce a novel training-free approach named DiffSegmenter. The insight is that to generate realistic objects that are semantically faithful to the input text, both the complete object shapes and the corresponding semantics are implicitly learned by diffusion models. We discover that the object shapes are characterized by the self-attention maps while the semantics are indicated through the cross-attention maps produced by the denoising U-Net, forming the basis of our segmentation results.Additionally, we carefully design effective textual prompts and a category filtering mechanism to further enhance the segmentation results. Extensive experiments on three benchmark datasets show that the proposed DiffSegmenter achieves impressive results for open-vocabulary semantic segmentation.
Authors: Wenlong Zhang, Xiaohui Li, Xiangyu Chen, Yu Qiao, Xiao-Ming Wu, Chao Dong
Real-world Super-Resolution (Real-SR) methods focus on dealing with diverse real-world images and have attracted increasing attention in recent years. The key idea is to use a complex and high-order degradation model to mimic real-world degradations. Although they have achieved impressive results in various scenarios, they are faced with the obstacle of evaluation. Currently, these methods are only assessed by their average performance on a small set of degradation cases randomly selected from a large space, which fails to provide a comprehensive understanding of their overall performance and often yields inconsistent and potentially misleading results. To overcome the limitation in evaluation, we propose SEAL, a framework for systematic evaluation of real-SR. In particular, we cluster the extensive degradation space to create a set of representative degradation cases, which serves as a comprehensive test set. Next, we propose a coarse-to-fine evaluation protocol to measure the distributed and relative performance of real-SR methods on the test set. The protocol incorporates two new metrics: acceptance rate (AR) and relative performance ratio (RPR), derived from acceptance and excellence lines. Under SEAL, we benchmark existing real-SR methods, obtain new observations and insights into their performance, and develop a new strong baseline. We consider SEAL as the first step towards creating a comprehensive real-SR evaluation platform, which can promote the development of real-SR. The source code is available at https://github.com/XPixelGroup/SEAL
Authors: Chen Zhao, Weiling Cai, Chenyu Dong, Ziqi Zeng
Underwater images suffer from complex and diverse degradation, which inevitably affects the performance of underwater visual tasks. However, most existing learning-based Underwater image enhancement (UIE) methods mainly restore such degradations in the spatial domain, and rarely pay attention to the fourier frequency information. In this paper, we develop a novel UIE framework based on spatial-frequency interaction and gradient maps, namely SFGNet, which consists of two stages. Specifically, in the first stage, we propose a dense spatial-frequency fusion network (DSFFNet), mainly including our designed dense fourier fusion block and dense spatial fusion block, achieving sufficient spatial-frequency interaction by cross connections between these two blocks. In the second stage, we propose a gradient-aware corrector (GAC) to further enhance perceptual details and geometric structures of images by gradient map. Experimental results on two real-world underwater image datasets show that our approach can successfully enhance underwater images, and achieves competitive performance in visual quality improvement. The code is available at https://github.com/zhihefang/SFGNet.
Authors: Zhengxiang Shi, Aldo Lipani
Prompt tuning (PT), where a small amount of trainable soft (continuous) prompt vectors is affixed to the input of language models (LM), has shown promising results across various tasks and models for parameter-efficient fine-tuning (PEFT). PT stands out from other PEFT approaches because it maintains competitive performance with fewer trainable parameters and does not drastically scale up its parameters as the model size expands. However, PT introduces additional soft prompt tokens, leading to longer input sequences, which significantly impacts training and inference time and memory usage due to the Transformer's quadratic complexity. Particularly concerning for Large Language Models (LLMs) that face heavy daily querying. To address this issue, we propose Decomposed Prompt Tuning (DePT), which decomposes the soft prompt into a shorter soft prompt and a pair of low-rank matrices that are then optimised with two different learning rates. This allows DePT to achieve better performance while saving substantial memory and time costs compared to vanilla PT and its variants, without changing trainable parameter sizes. Through extensive experiments on 23 natural language processing (NLP) and vision-language (VL) tasks, we demonstrate that DePT outperforms state-of-the-art PEFT approaches, including the full fine-tuning baseline, in some scenarios. Additionally, we empirically show that DEPT grows more efficient as the model size increases. Our further study reveals that DePT integrates seamlessly with parameter-efficient transfer learning in the few-shot learning setting and highlights its adaptability to various model architectures and sizes.
Authors: Qianliang Wu, Yaqing Ding, Lei Luo, Shuo Gu, Chuanwei Zhou, Jin Xie, Jian Yang
Point Cloud Registration (PCR) is a critical and challenging task in computer vision. One of the primary difficulties in PCR is identifying salient and meaningful points that exhibit consistent semantic and geometric properties across different scans. Previous methods have encountered challenges with ambiguous matching due to the similarity among patch blocks throughout the entire point cloud and the lack of consideration for efficient global geometric consistency. To address these issues, we propose a new framework that includes several novel techniques. Firstly, we introduce a semantic-aware geometric encoder that combines object-level and patch-level semantic information. This encoder significantly improves registration recall by reducing ambiguity in patch-level superpoint matching. Additionally, we incorporate a prior knowledge approach that utilizes an intrinsic shape signature to identify salient points. This enables us to extract the most salient super points and meaningful dense points in the scene. Secondly, we introduce an innovative transformer that encodes High-Order (HO) geometric features. These features are crucial for identifying salient points within initial overlap regions while considering global high-order geometric consistency. To optimize this high-order transformer further, we introduce an anchor node selection strategy. By encoding inter-frame triangle or polyhedron consistency features based on these anchor nodes, we can effectively learn high-order geometric features of salient super points. These high-order features are then propagated to dense points and utilized by a Sinkhorn matching module to identify key correspondences for successful registration. In our experiments conducted on well-known datasets such as 3DMatch/3DLoMatch and KITTI, our approach has shown promising results, highlighting the effectiveness of our novel method.
Authors: Xiaogang Jia, Songlei Jian, Yusong Tan, Yonggang Che, Wei Chen, Zhengfa Liang
Depth completion is a popular research direction in the field of depth estimation. The fusion of color and depth features is the current critical challenge in this task, mainly due to the asymmetry between the rich scene details in color images and the sparse pixels in depth maps. To tackle this issue, we design an efficient Gated Cross-Attention Network that propagates confidence via a gating mechanism, simultaneously extracting and refining key information in both color and depth branches to achieve local spatial feature fusion. Additionally, we employ an attention network based on the Transformer in low-dimensional space to effectively fuse global features and increase the network's receptive field. With a simple yet efficient gating mechanism, our proposed method achieves fast and accurate depth completion without the need for additional branches or post-processing steps. At the same time, we use the Ray Tune mechanism with the AsyncHyperBandScheduler scheduler and the HyperOptSearch algorithm to automatically search for the optimal number of module iterations, which also allows us to achieve performance comparable to state-of-the-art methods. We conduct experiments on both indoor and outdoor scene datasets. Our fast network achieves Pareto-optimal solutions in terms of time and accuracy, and at the time of submission, our accurate network ranks first among all published papers on the KITTI official website in terms of accuracy.
Authors: Tasin Islam, Alina Miron, XiaoHui Liu, Yongmin Li
Our study introduces a new image-to-video generator called FashionFlow to generate fashion videos. By utilising a diffusion model, we are able to create short videos from still fashion images. Our approach involves developing and connecting relevant components with the diffusion model, which results in the creation of high-fidelity videos that are aligned with the conditional image. The components include the use of pseudo-3D convolutional layers to generate videos efficiently. VAE and CLIP encoders capture vital characteristics from still images to condition the diffusion model at a global level. Our research demonstrates a successful synthesis of fashion videos featuring models posing from various angles, showcasing the fit and appearance of the garment. Our findings hold great promise for improving and enhancing the shopping experience for the online fashion industry.
Authors: Fadillah Maani, Asim Ukaye, Nada Saadi, Numan Saeed, Mohammad Yaqub
Echocardiography has become an indispensable clinical imaging modality for general heart health assessment. From calculating biomarkers such as ejection fraction to the probability of a patient's heart failure, accurate segmentation of the heart and its structures allows doctors to plan and execute treatments with greater precision and accuracy. However, achieving accurate and robust left ventricle segmentation is time-consuming and challenging due to different reasons. This work introduces a novel approach for consistent left ventricular (LV) segmentation from sparsely annotated echocardiogram videos. We achieve this through (1) self-supervised learning (SSL) using temporal masking followed by (2) weakly supervised training. We investigate two different segmentation approaches: 3D segmentation and a novel 2D superimage (SI). We demonstrate how our proposed method outperforms the state-of-the-art solutions by achieving a 93.32% (95%CI 93.21-93.43%) dice score on a large-scale dataset (EchoNet-Dynamic) while being more efficient. To show the effectiveness of our approach, we provide extensive ablation studies, including pre-training settings and various deep learning backbones. Additionally, we discuss how our proposed methodology achieves high data utility by incorporating unlabeled frames in the training process. To help support the AI in medicine community, the complete solution with the source code will be made publicly available upon acceptance.
Authors: Mustafa Shukor, Alexandre Rame, Corentin Dancette, Matthieu Cord
Following the success of Large Language Models (LLMs), Large Multimodal Models (LMMs), such as the Flamingo model and its subsequent competitors, have started to emerge as natural steps towards generalist agents. However, interacting with recent LMMs reveals major limitations that are hardly captured by the current evaluation benchmarks. Indeed, task performances (e.g., VQA accuracy) alone do not provide enough clues to understand their real capabilities, limitations, and to which extent such models are aligned to human expectations. To refine our understanding of those flaws, we deviate from the current evaluation paradigm, and (1) evaluate 10 recent open-source LMMs from 3B up to 80B parameter scale, on 5 different axes; hallucinations, abstention, compositionality, explainability and instruction following. Our evaluation on these axes reveals major flaws in LMMs. While the current go-to solution to align these models is based on training, such as instruction tuning or RLHF, we rather (2) explore the training-free in-context learning (ICL) as a solution, and study how it affects these limitations. Based on our ICL study, (3) we push ICL further and propose new multimodal ICL variants such as; Multitask-ICL, Chain-of-Hindsight-ICL, and Self-Correcting-ICL. Our findings are as follows. (1) Despite their success, LMMs have flaws that remain unsolved with scaling alone. (2) The effect of ICL on LMMs flaws is nuanced; despite its effectiveness for improved explainability, answer abstention, ICL only slightly improves instruction following, does not improve compositional abilities, and actually even amplifies hallucinations. (3) The proposed ICL variants are promising as post-hoc approaches to efficiently tackle some of those flaws. The code is available here: https://github.com/mshukor/EvALign-ICL.
Authors: Lirui Wang, Yiyang Ling, Zhecheng Yuan, Mohit Shridhar, Chen Bao, Yuzhe Qin, Bailin Wang, Huazhe Xu, Xiaolong Wang
Collecting large amounts of real-world interaction data to train general robotic policies is often prohibitively expensive, thus motivating the use of simulation data. However, existing methods for data generation have generally focused on scene-level diversity (e.g., object instances and poses) rather than task-level diversity, due to the human effort required to come up with and verify novel tasks. This has made it challenging for policies trained on simulation data to demonstrate significant task-level generalization. In this paper, we propose to automatically generate rich simulation environments and expert demonstrations by exploiting a large language models' (LLM) grounding and coding ability. Our approach, dubbed GenSim, has two modes: goal-directed generation, wherein a target task is given to the LLM and the LLM proposes a task curriculum to solve the target task, and exploratory generation, wherein the LLM bootstraps from previous tasks and iteratively proposes novel tasks that would be helpful in solving more complex tasks. We use GPT4 to expand the existing benchmark by ten times to over 100 tasks, on which we conduct supervised finetuning and evaluate several LLMs including finetuned GPTs and Code Llama on code generation for robotic simulation tasks. Furthermore, we observe that LLMs-generated simulation programs can enhance task-level generalization significantly when used for multitask policy training. We further find that with minimal sim-to-real adaptation, the multitask policies pretrained on GPT4-generated simulation tasks exhibit stronger transfer to unseen long-horizon tasks in the real world and outperform baselines by 25%. See the project website (https://liruiw.github.io/gensim) for code, demos, and videos.
Authors: Bin Zhu, Bin Lin, Munan Ning, Yang Yan, Jiaxi Cui, HongFa Wang, Yatian Pang, Wenhao Jiang, Junwu Zhang, Zongwei Li, Wancai Zhang, Zhifeng Li, Wei Liu, Li Yuan
The video-language (VL) pretraining has achieved remarkable improvement in multiple downstream tasks. However, the current VL pretraining framework is hard to extend to multiple modalities (N modalities, N>=3) beyond vision and language. We thus propose LanguageBind, taking the language as the bind across different modalities because the language modality is well-explored and contains rich semantics. Specifically, we freeze the language encoder acquired by VL pretraining, then train encoders for other modalities with contrastive learning. As a result, all modalities are mapped to a shared feature space, implementing multi-modal semantic alignment. While LanguageBind ensures that we can extend VL modalities to N modalities, we also need a high-quality dataset with alignment data pairs centered on language. We thus propose VIDAL-10M with Video, Infrared, Depth, Audio and their corresponding Language, naming as VIDAL-10M. In our VIDAL-10M, all videos are from short video platforms with complete semantics rather than truncated segments from long videos, and all the video, depth, infrared, and audio modalities are aligned to their textual descriptions. LanguageBind has achieved superior performance on a wide range of 15 benchmarks covering video, audio, depth, and infrared. Moreover, multiple experiments have provided evidence for the effectiveness of LanguageBind in achieving indirect alignment and complementarity among diverse modalities. Code address: https://github.com/PKU-YuanGroup/LanguageBind
Authors: Pan Lu, Hritik Bansal, Tony Xia, Jiacheng Liu, Chunyuan Li, Hannaneh Hajishirzi, Hao Cheng, Kai-Wei Chang, Michel Galley, Jianfeng Gao
Large Language Models (LLMs) and Large Multimodal Models (LMMs) exhibit impressive problem-solving skills in many tasks and domains, but their ability in mathematical reasoning in visual contexts has not been systematically studied. To bridge this gap, we present MathVista, a benchmark designed to combine challenges from diverse mathematical and visual tasks. It consists of 6,141 examples, derived from 28 existing multimodal datasets involving mathematics and 3 newly created datasets (i.e., IQTest, FunctionQA, and PaperQA). Completing these tasks requires fine-grained, deep visual understanding and compositional reasoning, which all state-of-the-art foundation models find challenging. With MathVista, we have conducted a comprehensive, quantitative evaluation of 12 prominent foundation models. The best-performing GPT-4V model achieves an overall accuracy of 49.9%, substantially outperforming Bard, the second-best performer, by 15.1%. Our in-depth analysis reveals that the superiority of GPT-4V is mainly attributed to its enhanced visual perception and mathematical reasoning. However, GPT-4V still falls short of human performance by 10.4%, as it often struggles to understand complex figures and perform rigorous reasoning. This significant gap underscores the critical role that MathVista will play in the development of general-purpose AI agents capable of tackling mathematically intensive and visually rich real-world tasks. We further explore the new ability of self-verification, the application of self-consistency, and the interactive chatbot capabilities of GPT-4V, highlighting its promising potential for future research. The project is available at https://mathvista.github.io/.
Authors: Yossi Gandelsman, Alexei A. Efros, Jacob Steinhardt
We investigate the CLIP image encoder by analyzing how individual model components affect the final representation. We decompose the image representation as a sum across individual image patches, model layers, and attention heads, and use CLIP's text representation to interpret the summands. Interpreting the attention heads, we characterize each head's role by automatically finding text representations that span its output space, which reveals property-specific roles for many heads (e.g. location or shape). Next, interpreting the image patches, we uncover an emergent spatial localization within CLIP. Finally, we use this understanding to remove spurious features from CLIP and to create a strong zero-shot image segmenter. Our results indicate that a scalable understanding of transformer models is attainable and can be used to repair and improve models.
Authors: Hansen Feng, Lizhi Wang, Yiqi Huang, Yuzhi Wang, Lin Zhu, Hua Huang
Recently, the mainstream practice for training low-light raw image denoising methods has shifted towards employing synthetic data. Noise modeling, which focuses on characterizing the noise distribution of real-world sensors, profoundly influences the effectiveness and practicality of synthetic data. Currently, physics-based noise modeling struggles to characterize the entire real noise distribution, while learning-based noise modeling impractically depends on paired real data. In this paper, we propose a novel strategy: learning the noise model from dark frames instead of paired real data, to break down the data dependency. Based on this strategy, we introduce an efficient physics-guided noise neural proxy (PNNP) to approximate the real-world sensor noise model. Specifically, we integrate physical priors into neural proxies and introduce three efficient techniques: physics-guided noise decoupling (PND), physics-guided proxy model (PPM), and differentiable distribution loss (DDL). PND decouples the dark frame into different components and handles different levels of noise flexibly, which reduces the complexity of noise modeling. PPM incorporates physical priors to constrain the generated noise, which promotes the accuracy of noise modeling. DDL provides explicit and reliable supervision for noise distribution, which promotes the precision of noise modeling. PNNP exhibits powerful potential in characterizing the real noise distribution. Extensive experiments on public datasets demonstrate superior performance in practical low-light raw image denoising. The code will be available at \url{https://github.com/fenghansen/PNNP}.
Authors: Xuewei Li, Yaqiao Zhu, Jie Gao, Xi Wei, Ruixuan Zhang, Yuan Tian, ZhiQiang Liu
Segmentation of nodules in thyroid ultrasound imaging plays a crucial role in the detection and treatment of thyroid cancer. However, owing to the diversity of scanner vendors and imaging protocols in different hospitals, the automatic segmentation model, which has already demonstrated expert-level accuracy in the field of medical image segmentation, finds its accuracy reduced as the result of its weak generalization performance when being applied in clinically realistic environments. To address this issue, the present paper proposes ASTN, a framework for thyroid nodule segmentation achieved through a new type co-registration network. By extracting latent semantic information from the atlas and target images and utilizing in-depth features to accomplish the co-registration of nodules in thyroid ultrasound images, this framework can ensure the integrity of anatomical structure and reduce the impact on segmentation as the result of overall differences in image caused by different devices. In addition, this paper also provides an atlas selection algorithm to mitigate the difficulty of co-registration. As shown by the evaluation results collected from the datasets of different devices, thanks to the method we proposed, the model generalization has been greatly improved while maintaining a high level of segmentation accuracy.
Authors: Cheng-Kun Yang, Min-Hung Chen, Yung-Yu Chuang, Yen-Yu Lin
We present a Multimodal Interlaced Transformer (MIT) that jointly considers 2D and 3D data for weakly supervised point cloud segmentation. Research studies have shown that 2D and 3D features are complementary for point cloud segmentation. However, existing methods require extra 2D annotations to achieve 2D-3D information fusion. Considering the high annotation cost of point clouds, effective 2D and 3D feature fusion based on weakly supervised learning is in great demand. To this end, we propose a transformer model with two encoders and one decoder for weakly supervised point cloud segmentation using only scene-level class tags. Specifically, the two encoders compute the self-attended features for 3D point clouds and 2D multi-view images, respectively. The decoder implements interlaced 2D-3D cross-attention and carries out implicit 2D and 3D feature fusion. We alternately switch the roles of queries and key-value pairs in the decoder layers. It turns out that the 2D and 3D features are iteratively enriched by each other. Experiments show that it performs favorably against existing weakly supervised point cloud segmentation methods by a large margin on the S3DIS and ScanNet benchmarks. The project page will be available at https://jimmy15923.github.io/mit_web/.
Authors: Yingshu Li, Yunyi Liu, Zhanyu Wang, Xinyu Liang, Lei Wang, Lingqiao Liu, Leyang Cui, Zhaopeng Tu, Longyue Wang, Luping Zhou
This work conducts an evaluation of GPT-4V's multimodal capability for medical image analysis, with a focus on three representative tasks of radiology report generation, medical visual question answering, and medical visual grounding. For the evaluation, a set of prompts is designed for each task to induce the corresponding capability of GPT-4V to produce sufficiently good outputs. Three evaluation ways including quantitative analysis, human evaluation, and case study are employed to achieve an in-depth and extensive evaluation. Our evaluation shows that GPT-4V excels in understanding medical images and is able to generate high-quality radiology reports and effectively answer questions about medical images. Meanwhile, it is found that its performance for medical visual grounding needs to be substantially improved. In addition, we observe the discrepancy between the evaluation outcome from quantitative analysis and that from human evaluation. This discrepancy suggests the limitations of conventional metrics in assessing the performance of large language models like GPT-4V and the necessity of developing new metrics for automatic quantitative analysis.
Authors: Yu Tian, Min Shi, Yan Luo, Ava Kouhana, Tobias Elze, Mengyu Wang
Fairness in artificial intelligence models has gained significantly more attention in recent years, especially in the area of medicine, as fairness in medical models is critical to people's well-being and lives. High-quality medical fairness datasets are needed to promote fairness learning research. Existing medical fairness datasets are all for classification tasks, and no fairness datasets are available for medical segmentation, while medical segmentation is an equally important clinical task as classifications, which can provide detailed spatial information on organ abnormalities ready to be assessed by clinicians. In this paper, we propose the first fairness dataset for medical segmentation named Harvard-FairSeg with 10,000 subject samples. In addition, we propose a fair error-bound scaling approach to reweight the loss function with the upper error-bound in each identity group, using the segment anything model (SAM). We anticipate that the segmentation performance equity can be improved by explicitly tackling the hard cases with high training errors in each identity group. To facilitate fair comparisons, we utilize a novel equity-scaled segmentation performance metric to compare segmentation metrics in the context of fairness, such as the equity-scaled Dice coefficient. Through comprehensive experiments, we demonstrate that our fair error-bound scaling approach either has superior or comparable fairness performance to the state-of-the-art fairness learning models. The dataset and code are publicly accessible via https://ophai.hms.harvard.edu/harvard-fairseg10k.
Authors: Elisa Warner, Joonsang Lee, William Hsu, Tanveer Syeda-Mahmood, Charles Kahn, Olivier Gevaert, Arvind Rao
Machine learning (ML) applications in medical artificial intelligence (AI) systems have shifted from traditional and statistical methods to increasing application of deep learning models. This survey navigates the current landscape of multimodal ML, focusing on its profound impact on medical image analysis and clinical decision support systems. Emphasizing challenges and innovations in addressing multimodal representation, fusion, translation, alignment, and co-learning, the paper explores the transformative potential of multimodal models for clinical predictions. It also highlights the need for principled assessments and practical implementation of such models, bringing attention to the dynamics between decision support systems and healthcare providers and personnel. Despite advancements, challenges such as data biases and the scarcity of "big data" in many biomedical domains persist. We conclude with a discussion on principled innovation and collaborative efforts to further the mission of seamless integration of multimodal ML models into biomedical practice.
Authors: Chenyu Gao, Michael E. Kim, Ho Hin Lee, Qi Yang, Nazirah Mohd Khairi, Praitayini Kanakaraj, Nancy R. Newlin, Derek B. Archer, Angela L. Jefferson, Warren D. Taylor, Brian D. Boyd, Lori L. Beason-Held, Susan M. Resnick, The BIOCARD Study Team, Yuankai Huo, Katherine D. Van Schaik, Kurt G. Schilling, Daniel Moyer, Ivana Išgum, Bennett A. Landman
Imaging findings inconsistent with those expected at specific chronological age ranges may serve as early indicators of neurological disorders and increased mortality risk. Estimation of chronological age, and deviations from expected results, from structural MRI data has become an important task for developing biomarkers that are sensitive to such deviations. Complementary to structural analysis, diffusion tensor imaging (DTI) has proven effective in identifying age-related microstructural changes within the brain white matter, thereby presenting itself as a promising additional modality for brain age prediction. Although early studies have sought to harness DTI's advantages for age estimation, there is no evidence that the success of this prediction is owed to the unique microstructural and diffusivity features that DTI provides, rather than the macrostructural features that are also available in DTI data. Therefore, we seek to develop white-matter-specific age estimation to capture deviations from normal white matter aging. Specifically, we deliberately disregard the macrostructural information when predicting age from DTI scalar images, using two distinct methods. The first method relies on extracting only microstructural features from regions of interest. The second applies 3D residual neural networks (ResNets) to learn features directly from the images, which are non-linearly registered and warped to a template to minimize macrostructural variations. When tested on unseen data, the first method yields mean absolute error (MAE) of 6.11 years for cognitively normal participants and MAE of 6.62 years for cognitively impaired participants, while the second method achieves MAE of 4.69 years for cognitively normal participants and MAE of 4.96 years for cognitively impaired participants. We find that the ResNet model captures subtler, non-macrostructural features for brain age prediction.
Authors: Wasim Ahmad, Yan-Tsung Peng, Yuan-Hao Chang, Gaddisa Olani Ganfure, Sarwar Khan, Sahibzada Adil Shahzad
Deepfake videos, generated through AI faceswapping techniques, have garnered considerable attention due to their potential for powerful impersonation attacks. While existing research primarily focuses on binary classification to discern between real and fake videos, however determining the specific generation model for a fake video is crucial for forensic investigation. Addressing this gap, this paper investigates the model attribution problem of Deepfake videos from a recently proposed dataset, Deepfakes from Different Models (DFDM), derived from various Autoencoder models. The dataset comprises 6,450 Deepfake videos generated by five distinct models with variations in encoder, decoder, intermediate layer, input resolution, and compression ratio. This study formulates Deepfakes model attribution as a multiclass classification task, proposing a segment of VGG19 as a feature extraction backbone, known for its effectiveness in imagerelated tasks, while integrated a Capsule Network with a Spatio-Temporal attention mechanism. The Capsule module captures intricate hierarchies among features for robust identification of deepfake attributes. Additionally, the video-level fusion technique leverages temporal attention mechanisms to handle concatenated feature vectors, capitalizing on inherent temporal dependencies in deepfake videos. By aggregating insights across frames, our model gains a comprehensive understanding of video content, resulting in more precise predictions. Experimental results on the deepfake benchmark dataset (DFDM) demonstrate the efficacy of our proposed method, achieving up to a 4% improvement in accurately categorizing deepfake videos compared to baseline models while demanding fewer computational resources.
Authors: Chayun Kongtongvattana, Baoru Huang, Jingxuan Kang, Hoan Nguyen, Olajide Olufemi, Anh Nguyen
We introduce a shape-sensitive loss function for catheter and guidewire segmentation and utilize it in a vision transformer network to establish a new state-of-the-art result on a large-scale X-ray images dataset. We transform network-derived predictions and their corresponding ground truths into signed distance maps, thereby enabling any networks to concentrate on the essential boundaries rather than merely the overall contours. These SDMs are subjected to the vision transformer, efficiently producing high-dimensional feature vectors encapsulating critical image attributes. By computing the cosine similarity between these feature vectors, we gain a nuanced understanding of image similarity that goes beyond the limitations of traditional overlap-based measures. The advantages of our approach are manifold, ranging from scale and translation invariance to superior detection of subtle differences, thus ensuring precise localization and delineation of the medical instruments within the images. Comprehensive quantitative and qualitative analyses substantiate the significant enhancement in performance over existing baselines, demonstrating the promise held by our new shape-sensitive loss function for improving catheter and guidewire segmentation.
Authors: Shuo Chen, Boxiao Liu, Haihang You
Spiking Neural Networks (SNNs) have gained considerable attention due to the energy-efficient and multiplication-free characteristics. The continuous growth in scale of deep SNNs poses challenges for model deployment. Network pruning reduces hardware resource requirements of model deployment by compressing the network scale. However, existing SNN pruning methods cause high pruning costs and performance loss because the pruning iterations amplify the training difficulty of SNNs. In this paper, inspired by the critical brain hypothesis in neuroscience, we propose a regeneration mechanism based on the neuron criticality for SNN pruning to enhance feature extraction and accelerate the pruning process. Firstly, we propose a low-cost metric for the criticality in SNNs. Then, we re-rank the pruned structures after pruning and regenerate those with higher criticality to obtain the critical network. Our method achieves higher performance than the current state-of-the-art (SOTA) method with up to 95.26% reduction of pruning cost. Moreover, we investigate the underlying mechanism of our method and find that it efficiently selects potential structures and learns the consistent feature representation.
Authors: Kunchang Li, Yali Wang, Yinan He, Yizhuo Li, Yi Wang, Yi Liu, Zun Wang, Jilan Xu, Guo Chen, Ping Luo, Limin Wang, Yu Qiao
With the rapid development of Multi-modal Large Language Models (MLLMs), a number of diagnostic benchmarks have recently emerged to evaluate the comprehension capabilities of these models. However, most benchmarks predominantly assess spatial understanding in the static image tasks, while overlooking temporal understanding in the dynamic video tasks. To alleviate this issue, we introduce a comprehensive Multi-modal Video understanding Benchmark, namely MVBench, which covers 20 challenging video tasks that cannot be effectively solved with a single frame. Specifically, we first introduce a novel static-to-dynamic method to define these temporal-related tasks. By transforming various static tasks into dynamic ones, we enable the systematic generation of video tasks that require a broad spectrum of temporal skills, ranging from perception to cognition. Then, guided by the task definition, we automatically convert public video annotations into multiple-choice QA to evaluate each task. On one hand, such a distinct paradigm allows us to build MVBench efficiently, without much manual intervention. On the other hand, it guarantees evaluation fairness with ground-truth video annotations, avoiding the biased scoring of LLMs. Moreover, we further develop a robust video MLLM baseline, i.e., VideoChat2, by progressive multi-modal training with diverse instruction-tuning data. The extensive results on our MVBench reveal that, the existing MLLMs are far from satisfactory in temporal understanding, while our VideoChat2 largely surpasses these leading models by over 15% on MVBench. All models and data are available at https://github.com/OpenGVLab/Ask-Anything.
Authors: Benjamin Schneider, Nils Lukas, Florian Kerschbaum
Web-scraped datasets are vulnerable to data poisoning, which can be used for backdooring deep image classifiers during training. Since training on large datasets is expensive, a model is trained once and re-used many times. Unlike adversarial examples, backdoor attacks often target specific classes rather than any class learned by the model. One might expect that targeting many classes through a naive composition of attacks vastly increases the number of poison samples. We show this is not necessarily true and more efficient, universal data poisoning attacks exist that allow controlling misclassifications from any source class into any target class with a small increase in poison samples. Our idea is to generate triggers with salient characteristics that the model can learn. The triggers we craft exploit a phenomenon we call inter-class poison transferability, where learning a trigger from one class makes the model more vulnerable to learning triggers for other classes. We demonstrate the effectiveness and robustness of our universal backdoor attacks by controlling models with up to 6,000 classes while poisoning only 0.15% of the training dataset. Our source code is available at https://github.com/Ben-Schneider-code/Universal-Backdoor-Attacks.
Authors: Jie Wang, Jiu-Cheng Xie, Xianyan Li, Feng Xu, Chi-Man Pun, Hao Gao
Constructing vivid 3D head avatars for given subjects and realizing a series of animations on them is valuable yet challenging. This paper presents GaussianHead, which models the actional human head with anisotropic 3D Gaussians. In our framework, a motion deformation field and multi-resolution tri-plane are constructed respectively to deal with the head's dynamic geometry and complex texture. Notably, we impose an exclusive derivation scheme on each Gaussian, which generates its multiple doppelgangers through a set of learnable parameters for position transformation. With this design, we can compactly and accurately encode the appearance information of Gaussians, even those fitting the head's particular components with sophisticated structures. In addition, an inherited derivation strategy for newly added Gaussians is adopted to facilitate training acceleration. Extensive experiments show that our method can produce high-fidelity renderings, outperforming state-of-the-art approaches in reconstruction, cross-identity reenactment, and novel view synthesis tasks. Our code is available at: https://github.com/chiehwangs/gaussian-head.
Authors: Minhyun Lee, Song Park, Byeongho Heo, Dongyoon Han, Hyunjung Shim
Recent advancements in Deep Neural Network (DNN) models have significantly improved performance across computer vision tasks. However, achieving highly generalizable and high-performing vision models requires extensive datasets, leading to large storage requirements. This storage challenge poses a critical bottleneck for scaling up vision models. Motivated by the success of discrete representations, SeiT proposes to use Vector-Quantized (VQ) feature vectors (i.e., tokens) as network inputs for vision classification. However, applying traditional data augmentations to tokens faces challenges due to input domain shift. To address this issue, we introduce TokenAdapt and ColorAdapt, simple yet effective token-based augmentation strategies. TokenAdapt realigns token embedding space for compatibility with spatial augmentations, preserving the model's efficiency without requiring fine-tuning. Additionally, ColorAdapt addresses color-based augmentations for tokens inspired by Adaptive Instance Normalization (AdaIN). We evaluate our approach across various scenarios, including storage-efficient ImageNet-1k classification, fine-grained classification, robustness benchmarks, and ADE-20k semantic segmentation. Experimental results demonstrate consistent performance improvement in diverse experiments. Code is available at https://github.com/naver-ai/tokenadapt.
Authors: Chaojian Li, Bichen Wu, Peter Vajda, Yingyan (Celine)Lin
Neural Radiance Field (NeRF) has emerged as a leading technique for novel view synthesis, owing to its impressive photorealistic reconstruction and rendering capability. Nevertheless, achieving real-time NeRF rendering in large-scale scenes has presented challenges, often leading to the adoption of either intricate baked mesh representations with a substantial number of triangles or resource-intensive ray marching in baked representations. We challenge these conventions, observing that high-quality geometry, represented by meshes with substantial triangles, is not necessary for achieving photorealistic rendering quality. Consequently, we propose MixRT, a novel NeRF representation that includes a low-quality mesh, a view-dependent displacement map, and a compressed NeRF model. This design effectively harnesses the capabilities of existing graphics hardware, thus enabling real-time NeRF rendering on edge devices. Leveraging a highly-optimized WebGL-based rendering framework, our proposed MixRT attains real-time rendering speeds on edge devices (over 30 FPS at a resolution of 1280 x 720 on a MacBook M1 Pro laptop), better rendering quality (0.2 PSNR higher in indoor scenes of the Unbounded-360 datasets), and a smaller storage size (less than 80% compared to state-of-the-art methods).
Authors: Andong Lu, Jiacong Zhao, Chenglong Li, Jin Tang, Bin Luo
Current RGBT tracking researches mainly focus on the modality-complete scenarios, overlooking the modality-missing challenge in real-world scenes. In this work, we comprehensively investigate the impact of modality-missing challenge in RGBT tracking and propose a novel invertible prompt learning approach, which integrates the content-preserving prompts into a well-trained tracking model to adapt to various modality-missing scenarios, for modality-missing RGBT tracking. In particular, given one modality-missing scenario, we propose to utilize the available modality to generate the prompt of the missing modality to adapt to RGBT tracking model. However, the cross-modality gap between available and missing modalities usually causes semantic distortion and information loss in prompt generation. To handle this issue, we propose the invertible prompt learning scheme by incorporating the full reconstruction of the input available modality from the prompt in prompt generation model. Considering that there lacks a modality-missing RGBT tracking dataset and many modality-missing scenarios are difficult to capture, we design a high-quality data simulation method based on hierarchical combination schemes to generate real-world modality-missing data. Extensive experiments on three modality-missing datasets show that our method achieves significant performance improvements compared with state-of-the-art methods. We will release the code and simulation dataset.
Authors: Pratibha Kumari, Joohi Chauhan, Afshin Bozorgpour, Boqiang Huang, Reza Azad, Dorit Merhof
Medical imaging analysis has witnessed remarkable advancements even surpassing human-level performance in recent years, driven by the rapid development of advanced deep-learning algorithms. However, when the inference dataset slightly differs from what the model has seen during one-time training, the model performance is greatly compromised. The situation requires restarting the training process using both the old and the new data which is computationally costly, does not align with the human learning process, and imposes storage constraints and privacy concerns. Alternatively, continual learning has emerged as a crucial approach for developing unified and sustainable deep models to deal with new classes, tasks, and the drifting nature of data in non-stationary environments for various application areas. Continual learning techniques enable models to adapt and accumulate knowledge over time, which is essential for maintaining performance on evolving datasets and novel tasks. This systematic review paper provides a comprehensive overview of the state-of-the-art in continual learning techniques applied to medical imaging analysis. We present an extensive survey of existing research, covering topics including catastrophic forgetting, data drifts, stability, and plasticity requirements. Further, an in-depth discussion of key components of a continual learning framework such as continual learning scenarios, techniques, evaluation schemes, and metrics is provided. Continual learning techniques encompass various categories, including rehearsal, regularization, architectural, and hybrid strategies. We assess the popularity and applicability of continual learning categories in various medical sub-fields like radiology and histopathology...
Authors: Senqiao Yang, Tianyuan Qu, Xin Lai, Zhuotao Tian, Bohao Peng, Shu Liu, Jiaya Jia
While LISA effectively bridges the gap between segmentation and large language models to enable reasoning segmentation, it poses certain limitations: unable to distinguish different instances of the target region, and constrained by the pre-defined textual response formats. In this work, we introduce LISA++, an update to the existing LISA model, focusing on improving core functionalities while keeping the base architecture intact. The main enhancements in LISA++ include: \textbf{1) Enhanced Segmentation}: The instance segmentation ability has been added, providing a more detailed scene analysis along with the existing multi-region semantic segmentation. \textbf{2) More Natural Conversation}: Improved capability for multi-turn dialogue, with the ability to incorporate segmentation results directly into text responses, i.e., Segmentation in Dialogue (SiD). These improvements are achieved by curating the existing samples of generic segmentation datasets, aimed specifically at enhancing the segmentation and conversational skills without structural change and additional data sources. Comparative analysis with the original LISA model shows significant advancements in these areas, positioning LISA++ as a notable upgrade in visual understanding and interaction. LISA++'s adaptability and improved features highlight the versatility of the mask-as-embedding paradigm proposed by LISA, and the potential as a foundational model for diverse applications.
Authors: Sebastian-Vasile Echim, Iulian-Marius Tăiatu, Dumitru-Clementin Cercel, Florin Pop
This work focuses on plant leaf disease classification and explores three crucial aspects: adversarial training, model explainability, and model compression. The models' robustness against adversarial attacks is enhanced through adversarial training, ensuring accurate classification even in the presence of threats. Leveraging explainability techniques, we gain insights into the model's decision-making process, improving trust and transparency. Additionally, we explore model compression techniques to optimize computational efficiency while maintaining classification performance. Through our experiments, we determine that on a benchmark dataset, the robustness can be the price of the classification accuracy with performance reductions of 3%-20% for regular tests and gains of 50%-70% for adversarial attack tests. We also demonstrate that a student model can be 15-25 times more computationally efficient for a slight performance reduction, distilling the knowledge of more complex models.
Authors: Simon Niedermayr, Josef Stumpfegger, Rüdiger Westermann
Recently, high-fidelity scene reconstruction with an optimized 3D Gaussian splat representation has been introduced for novel view synthesis from sparse image sets. Making such representations suitable for applications like network streaming and rendering on low-power devices requires significantly reduced memory consumption as well as improved rendering efficiency. We propose a compressed 3D Gaussian splat representation that utilizes sensitivity-aware vector clustering with quantization-aware training to compress directional colors and Gaussian parameters. The learned codebooks have low bitrates and achieve a compression rate of up to $31\times$ on real-world scenes with only minimal degradation of visual quality. We demonstrate that the compressed splat representation can be efficiently rendered with hardware rasterization on lightweight GPUs at up to $4\times$ higher framerates than reported via an optimized GPU compute pipeline. Extensive experiments across multiple datasets demonstrate the robustness and rendering speed of the proposed approach.
Authors: Jincen Jiang, Lizhi Zhao, Xuequan Lu, Wei Hu, Imran Razzak, Meili Wang
Recent works attempt to extend Graph Convolution Networks (GCNs) to point clouds for classification and segmentation tasks. These works tend to sample and group points to create smaller point sets locally and mainly focus on extracting local features through GCNs, while ignoring the relationship between point sets. In this paper, we propose the Dynamic Hop Graph Convolution Network (DHGCN) for explicitly learning the contextual relationships between the voxelized point parts, which are treated as graph nodes. Motivated by the intuition that the contextual information between point parts lies in the pairwise adjacent relationship, which can be depicted by the hop distance of the graph quantitatively, we devise a novel self-supervised part-level hop distance reconstruction task and design a novel loss function accordingly to facilitate training. In addition, we propose the Hop Graph Attention (HGA), which takes the learned hop distance as input for producing attention weights to allow edge features to contribute distinctively in aggregation. Eventually, the proposed DHGCN is a plug-and-play module that is compatible with point-based backbone networks. Comprehensive experiments on different backbones and tasks demonstrate that our self-supervised method achieves state-of-the-art performance. Our source code is available at: https://github.com/Jinec98/DHGCN.
Authors: Ziyue Huang, Mingming Zhang, Yuan Gong, Qingjie Liu, Yunhong Wang
Deep learning models are essential for scene classification, change detection, land cover segmentation, and other remote sensing image understanding tasks. Most backbones of existing remote sensing deep learning models are typically initialized by pre-trained weights obtained from ImageNet pre-training (IMP). However, domain gaps exist between remote sensing images and natural images (e.g., ImageNet), making deep learning models initialized by pre-trained weights of IMP perform poorly for remote sensing image understanding. Although some pre-training methods are studied in the remote sensing community, current remote sensing pre-training methods face the problem of vague generalization by only using remote sensing images. In this paper, we propose a novel remote sensing pre-training framework, Generic Knowledge Boosted Remote Sensing Pre-training (GeRSP), to learn robust representations from remote sensing and natural images for remote sensing understanding tasks. GeRSP contains two pre-training branches: (1) A self-supervised pre-training branch is adopted to learn domain-related representations from unlabeled remote sensing images. (2) A supervised pre-training branch is integrated into GeRSP for general knowledge learning from labeled natural images. Moreover, GeRSP combines two pre-training branches using a teacher-student architecture to simultaneously learn representations with general and special knowledge, which generates a powerful pre-trained model for deep learning model initialization. Finally, we evaluate GeRSP and other remote sensing pre-training methods on three downstream tasks, i.e., object detection, semantic segmentation, and scene classification. The extensive experimental results consistently demonstrate that GeRSP can effectively learn robust representations in a unified manner, improving the performance of remote sensing downstream tasks.
Authors: Alex Havrilla, Kevin Rojas, Wenjing Liao, Molei Tao
Diffusion generative models have achieved remarkable success in generating images with a fixed resolution. However, existing models have limited ability to generalize to different resolutions when training data at those resolutions are not available. Leveraging techniques from operator learning, we present a novel deep-learning architecture, Dual-FNO UNet (DFU), which approximates the score operator by combining both spatial and spectral information at multiple resolutions. Comparisons of DFU to baselines demonstrate its scalability: 1) simultaneously training on multiple resolutions improves FID over training at any single fixed resolution; 2) DFU generalizes beyond its training resolutions, allowing for coherent, high-fidelity generation at higher-resolutions with the same model, i.e. zero-shot super-resolution image-generation; 3) we propose a fine-tuning strategy to further enhance the zero-shot super-resolution image-generation capability of our model, leading to a FID of 11.3 at 1.66 times the maximum training resolution on FFHQ, which no other method can come close to achieving.
Authors: Vinh Quoc Luu, Duy Khanh Le, Huy Thanh Nguyen, Minh Thanh Nguyen, Thinh Tien Nguyen, Vinh Quang Dinh
Artificial Intelligence (AI) in healthcare, especially in white blood cell cancer diagnosis, is hindered by two primary challenges: the lack of large-scale labeled datasets for white blood cell (WBC) segmentation and outdated segmentation methods. To address the first challenge, a semi-supervised learning framework should be brought to efficiently annotate the large dataset. In this work, we address this issue by proposing a novel self-training pipeline with the incorporation of FixMatch. We discover that by incorporating FixMatch in the self-training pipeline, the performance improves in the majority of cases. Our performance achieved the best performance with the self-training scheme with consistency on DeepLab-V3 architecture and ResNet-50, reaching 90.69%, 87.37%, and 76.49% on Zheng 1, Zheng 2, and LISC datasets, respectively.
Authors: Guangyu Meng, Ruyu Zhou, Liu Liu, Peixian Liang, Fang Liu, Danny Chen, Michael Niemier, X.Sharon Hu
Earth Mover's Distance (EMD) is an important similarity measure between two distributions, used in computer vision and many other application domains. However, its exact calculation is computationally and memory intensive, which hinders its scalability and applicability for large-scale problems. Various approximate EMD algorithms have been proposed to reduce computational costs, but they suffer lower accuracy and may require additional memory usage or manual parameter tuning. In this paper, we present a novel approach, NNS-EMD, to approximate EMD using Nearest Neighbor Search (NNS), in order to achieve high accuracy, low time complexity, and high memory efficiency. The NNS operation reduces the number of data points compared in each NNS iteration and offers opportunities for parallel processing. We further accelerate NNS-EMD via vectorization on GPU, which is especially beneficial for large datasets. We compare NNS-EMD with both the exact EMD and state-of-the-art approximate EMD algorithms on image classification and retrieval tasks. We also apply NNS-EMD to calculate transport mapping and realize color transfer between images. NNS-EMD can be 44x to 135x faster than the exact EMD implementation, and achieves superior accuracy, speedup, and memory efficiency over existing approximate EMD methods.
Authors: Zhifeng Xie, Hao Li, Huiming Ding, Mengtian Li, Ying Cao
Cross-modal fashion synthesis and editing offer intelligent support to fashion designers by enabling the automatic generation and local modification of design drafts.While current diffusion models demonstrate commendable stability and controllability in image synthesis,they still face significant challenges in generating fashion design from abstract design elements and fine-grained editing.Abstract sensory expressions, \eg office, business, and party, form the high-level design concepts, while measurable aspects like sleeve length, collar type, and pant length are considered the low-level attributes of clothing.Controlling and editing fashion images using lengthy text descriptions poses a difficulty.In this paper, we propose HieraFashDiff,a novel fashion design method using the shared multi-stage diffusion model encompassing high-level design concepts and low-level clothing attributes in a hierarchical structure.Specifically, we categorized the input text into different levels and fed them in different time step to the diffusion model according to the criteria of professional clothing designers.HieraFashDiff allows designers to add low-level attributes after high-level prompts for interactive editing incrementally.In addition, we design a differentiable loss function in the sampling process with a mask to keep non-edit areas.Comprehensive experiments performed on our newly conducted Hierarchical fashion dataset,demonstrate that our proposed method outperforms other state-of-the-art competitors.
Authors: Zhenhui Ye, Tianyun Zhong, Yi Ren, Jiaqi Yang, Weichuang Li, Jiawei Huang, Ziyue Jiang, Jinzheng He, Rongjie Huang, Jinglin Liu, Chen Zhang, Xiang Yin, Zejun Ma, Zhou Zhao
One-shot 3D talking portrait generation aims to reconstruct a 3D avatar from an unseen image, and then animate it with a reference video or audio to generate a talking portrait video. The existing methods fail to simultaneously achieve the goals of accurate 3D avatar reconstruction and stable talking face animation. Besides, while the existing works mainly focus on synthesizing the head part, it is also vital to generate natural torso and background segments to obtain a realistic talking portrait video. To address these limitations, we present Real3D-Potrait, a framework that (1) improves the one-shot 3D reconstruction power with a large image-to-plane model that distills 3D prior knowledge from a 3D face generative model; (2) facilitates accurate motion-conditioned animation with an efficient motion adapter; (3) synthesizes realistic video with natural torso movement and switchable background using a head-torso-background super-resolution model; and (4) supports one-shot audio-driven talking face generation with a generalizable audio-to-motion model. Extensive experiments show that Real3D-Portrait generalizes well to unseen identities and generates more realistic talking portrait videos compared to previous methods. Video samples and source code are available at https://real3dportrait.github.io .
Authors: Bang An, Mucong Ding, Tahseen Rabbani, Aakriti Agrawal, Yuancheng Xu, Chenghao Deng, Sicheng Zhu, Abdirisak Mohamed, Yuxin Wen, Tom Goldstein, Furong Huang
This paper investigates the weaknesses of image watermarking techniques. We present WAVES (Watermark Analysis Via Enhanced Stress-testing), a novel benchmark for assessing watermark robustness, overcoming the limitations of current evaluation methods.WAVES integrates detection and identification tasks, and establishes a standardized evaluation protocol comprised of a diverse range of stress tests. The attacks in WAVES range from traditional image distortions to advanced and novel variations of diffusive, and adversarial attacks. Our evaluation examines two pivotal dimensions: the degree of image quality degradation and the efficacy of watermark detection after attacks. We develop a series of Performance vs. Quality 2D plots, varying over several prominent image similarity metrics, which are then aggregated in a heuristically novel manner to paint an overall picture of watermark robustness and attack potency. Our comprehensive evaluation reveals previously undetected vulnerabilities of several modern watermarking algorithms. We envision WAVES as a toolkit for the future development of robust watermarking systems. The project is available at https://wavesbench.github.io/
Authors: Nicholas Konz, Maciej A. Mazurowski
This paper investigates discrepancies in how neural networks learn from different imaging domains, which are commonly overlooked when adopting computer vision techniques from the domain of natural images to other specialized domains such as medical images. Recent works have found that the generalization error of a trained network typically increases with the intrinsic dimension ($d_{data}$) of its training set. Yet, the steepness of this relationship varies significantly between medical (radiological) and natural imaging domains, with no existing theoretical explanation. We address this gap in knowledge by establishing and empirically validating a generalization scaling law with respect to $d_{data}$, and propose that the substantial scaling discrepancy between the two considered domains may be at least partially attributed to the higher intrinsic "label sharpness" ($K_F$) of medical imaging datasets, a metric which we propose. Next, we demonstrate an additional benefit of measuring the label sharpness of a training set: it is negatively correlated with the trained model's adversarial robustness, which notably leads to models for medical images having a substantially higher vulnerability to adversarial attack. Finally, we extend our $d_{data}$ formalism to the related metric of learned representation intrinsic dimension ($d_{repr}$), derive a generalization scaling law with respect to $d_{repr}$, and show that $d_{data}$ serves as an upper bound for $d_{repr}$. Our theoretical results are supported by thorough experiments with six models and eleven natural and medical imaging datasets over a range of training set sizes. Our findings offer insights into the influence of intrinsic dataset properties on generalization, representation learning, and robustness in deep neural networks.
Authors: Tudor Jianu, Baoru Huang, Tuan Vo, Minh Nhat Vu, Jingxuan Kang, Hoan Nguyen, Olatunji Omisore, Pierre Berthet-Rayne, Sebastiano Fichera, Anh Nguyen
Endovascular robots have been actively developed in both academia and industry. However, progress toward autonomous catheterization is often hampered by the widespread use of closed-source simulators and physical phantoms. Additionally, the acquisition of large-scale datasets for training machine learning algorithms with endovascular robots is usually infeasible due to expensive medical procedures. In this chapter, we introduce CathSim, the first open-source simulator for endovascular intervention to address these limitations. CathSim emphasizes real-time performance to enable rapid development and testing of learning algorithms. We validate CathSim against the real robot and show that our simulator can successfully mimic the behavior of the real robot. Based on CathSim, we develop a multimodal expert navigation network and demonstrate its effectiveness in downstream endovascular navigation tasks. The intensive experimental results suggest that CathSim has the potential to significantly accelerate research in the autonomous catheterization field. Our project is publicly available at https://github.com/airvlab/cathsim.
Authors: Win Kent Ong, Kam Woh Ng, Chee Seng Chan, Yi Zhe Song, Tao Xiang
Neural Radiance Field (NeRF) models have gained significant attention in the computer vision community in the recent past with state-of-the-art visual quality and produced impressive demonstrations. Since then, technopreneurs have sought to leverage NeRF models into a profitable business. Therefore, NeRF models make it worth the risk of plagiarizers illegally copying, re-distributing, or misusing those models. This paper proposes a comprehensive intellectual property (IP) protection framework for the NeRF model in both black-box and white-box settings, namely IPR-NeRF. In the black-box setting, a diffusion-based solution is introduced to embed and extract the watermark via a two-stage optimization process. In the white-box setting, a designated digital signature is embedded into the weights of the NeRF model by adopting the sign loss objective. Our extensive experiments demonstrate that not only does our approach maintain the fidelity (\ie, the rendering quality) of IPR-NeRF models, but it is also robust against both ambiguity and removal attacks compared to prior arts.
Authors: Jianan Fan, Dongnan Liu, Hang Chang, Weidong Cai
Annotation scarcity and cross-modality/stain data distribution shifts are two major obstacles hindering the application of deep learning models for nuclei analysis, which holds a broad spectrum of potential applications in digital pathology. Recently, unsupervised domain adaptation (UDA) methods have been proposed to mitigate the distributional gap between different imaging modalities for unsupervised nuclei segmentation in histopathology images. However, existing UDA methods are built upon the assumption that data distributions within each domain should be uniform. Based on the over-simplified supposition, they propose to align the histopathology target domain with the source domain integrally, neglecting severe intra-domain discrepancy over subpartitions incurred by mixed cancer types and sampling organs. In this paper, for the first time, we propose to explicitly consider the heterogeneity within the histopathology domain and introduce open compound domain adaptation (OCDA) to resolve the crux. In specific, a two-stage disentanglement framework is proposed to acquire domain-invariant feature representations at both image and instance levels. The holistic design addresses the limitations of existing OCDA approaches which struggle to capture instance-wise variations. Two regularization strategies are specifically devised herein to leverage the rich subpartition-specific characteristics in histopathology images and facilitate subdomain decomposition. Moreover, we propose a dual-branch nucleus shape and structure preserving module to prevent nucleus over-generation and deformation in the synthesized images. Experimental results on both cross-modality and cross-stain scenarios over a broad range of diverse datasets demonstrate the superiority of our method compared with state-of-the-art UDA and OCDA methods.
Authors: Sagar Shrestha, Xiao Fu
Unsupervised domain translation (UDT) aims to find functions that convert samples from one domain (e.g., sketches) to another domain (e.g., photos) without changing the high-level semantic meaning (also referred to as ``content''). The translation functions are often sought by probability distribution matching of the transformed source domain and target domain. CycleGAN stands as arguably the most representative approach among this line of work. However, it was noticed in the literature that CycleGAN and variants could fail to identify the desired translation functions and produce content-misaligned translations. This limitation arises due to the presence of multiple translation functions -- referred to as ``measure-preserving automorphism" (MPA) -- in the solution space of the learning criteria. Despite awareness of such identifiability issues, solutions have remained elusive. This study delves into the core identifiability inquiry and introduces an MPA elimination theory. Our analysis shows that MPA is unlikely to exist, if multiple pairs of diverse cross-domain conditional distributions are matched by the learning function. Our theory leads to a UDT learner using distribution matching over auxiliary variable-induced subsets of the domains -- other than over the entire data domains as in the classical approaches. The proposed framework is the first to rigorously establish translation identifiability under reasonable UDT settings, to our best knowledge. Experiments corroborate with our theoretical claims.
Authors: Chen-Bin Feng, Qi Lai, Kangdao Liu, Houcheng Su, Chi-Man Vong
In semantic segmentation, accurate prediction masks are crucial for downstream tasks such as medical image analysis and image editing. Due to the lack of annotated data, few-shot semantic segmentation (FSS) performs poorly in predicting masks with precise contours. Recently, we have noticed that the large foundation model segment anything model (SAM) performs well in processing detailed features. Inspired by SAM, we propose FSS-SAM to boost FSS methods by addressing the issue of inaccurate contour. The FSS-SAM is training-free. It works as a post-processing tool for any FSS methods and can improve the accuracy of predicted masks. Specifically, we use predicted masks from FSS methods to generate prompts and then use SAM to predict new masks. To avoid predicting wrong masks with SAM, we propose a prediction result selection (PRS) algorithm. The algorithm can remarkably decrease wrong predictions. Experiment results on public datasets show that our method is superior to base FSS methods in both quantitative and qualitative aspects.
Authors: Changgu Chen, Junwei Shu, Lianggangxu Chen, Gaoqi He, Changbo Wang, Yang Li
Recent large-scale pre-trained diffusion models have demonstrated a powerful generative ability to produce high-quality videos from detailed text descriptions. However, exerting control over the motion of objects in videos generated by any video diffusion model is a challenging problem. In this paper, we propose a novel zero-shot moving object trajectory control framework, Motion-Zero, to enable a bounding-box-trajectories-controlled text-to-video diffusion model. To this end, an initial noise prior module is designed to provide a position-based prior to improve the stability of the appearance of the moving object and the accuracy of position. In addition, based on the attention map of the U-net, spatial constraints are directly applied to the denoising process of diffusion models, which further ensures the positional and spatial consistency of moving objects during the inference. Furthermore, temporal consistency is guaranteed with a proposed shift temporal attention mechanism. Our method can be flexibly applied to various state-of-the-art video diffusion models without any training process. Extensive experiments demonstrate our proposed method can control the motion trajectories of objects and generate high-quality videos.
Authors: Ling Yang, Jingwei Liu, Shenda Hong, Zhilong Zhang, Zhilin Huang, Zheming Cai, Wentao Zhang, Bin Cui
Diffusion models are a new class of generative models, and have dramatically promoted image generation with unprecedented quality and diversity. Existing diffusion models mainly try to reconstruct input image from a corrupted one with a pixel-wise or feature-wise constraint along spatial axes. However, such point-based reconstruction may fail to make each predicted pixel/feature fully preserve its neighborhood context, impairing diffusion-based image synthesis. As a powerful source of automatic supervisory signal, context has been well studied for learning representations. Inspired by this, we for the first time propose ConPreDiff to improve diffusion-based image synthesis with context prediction. We explicitly reinforce each point to predict its neighborhood context (i.e., multi-stride features/tokens/pixels) with a context decoder at the end of diffusion denoising blocks in training stage, and remove the decoder for inference. In this way, each point can better reconstruct itself by preserving its semantic connections with neighborhood context. This new paradigm of ConPreDiff can generalize to arbitrary discrete and continuous diffusion backbones without introducing extra parameters in sampling procedure. Extensive experiments are conducted on unconditional image generation, text-to-image generation and image inpainting tasks. Our ConPreDiff consistently outperforms previous methods and achieves a new SOTA text-to-image generation results on MS-COCO, with a zero-shot FID score of 6.21.
Authors: Weizheng Wang, Le Mao, Baijian Yang, Guohua Chen, Byung-Cheol Min
Predicting crowded intents and trajectories is crucial in varouls real-world applications, including service robots and autonomous vehicles. Understanding environmental dynamics is challenging, not only due to the complexities of modeling pair-wise spatial and temporal interactions but also the diverse influence of group-wise interactions. To decode the comprehensive pair-wise and group-wise interactions in crowded scenarios, we introduce Hyper-STTN, a Hypergraph-based Spatial-Temporal Transformer Network for crowd trajectory prediction. In Hyper-STTN, crowded group-wise correlations are constructed using a set of multi-scale hypergraphs with varying group sizes, captured through random-walk robability-based hypergraph spectral convolution. Additionally, a spatial-temporal transformer is adapted to capture pedestrians' pair-wise latent interactions in spatial-temporal dimensions. These heterogeneous group-wise and pair-wise are then fused and aligned though a multimodal transformer network. Hyper-STTN outperformes other state-of-the-art baselines and ablation models on 5 real-world pedestrian motion datasets.
Authors: Ping Guo, Zhiyuan Yang, Xi Lin, Qingchuan Zhao, Qingfu Zhang
Black-box query-based attacks constitute significant threats to Machine Learning as a Service (MLaaS) systems since they can generate adversarial examples without accessing the target model's architecture and parameters. Traditional defense mechanisms, such as adversarial training, gradient masking, and input transformations, either impose substantial computational costs or compromise the test accuracy of non-adversarial inputs. To address these challenges, we propose an efficient defense mechanism, PuriDefense, that employs random patch-wise purifications with an ensemble of lightweight purification models at a low level of inference cost. These models leverage the local implicit function and rebuild the natural image manifold. Our theoretical analysis suggests that this approach slows down the convergence of query-based attacks by incorporating randomness into purifications. Extensive experiments on CIFAR-10 and ImageNet validate the effectiveness of our proposed purifier-based defense mechanism, demonstrating significant improvements in robustness against query-based attacks.