... IEEE International Conference on Computer Vision workshops. IEEE International Conference on Computer Vision最新文献

Recent advances in text-to-image diffusion models have been driven by the increasing availability of paired 2D data. However, the development of 3D diffusion models has been hindered by the scarcity of high-quality 3D data, resulting in less competitive performance compared to their 2D counterparts. To address this challenge, we propose repurposing pre-trained 2D diffusion models for 3D object generation. We introduce Gaussian Atlas, a novel representation that utilizes dense 2D grids, enabling the fine-tuning of 2D diffusion models to generate 3D Gaussians. Our approach demonstrates successful transfer learning from a pre-trained 2D diffusion model to a 2D manifold flattened from 3D structures. To support model training, we compile GaussianVerse, a large-scale dataset comprising 205K high-quality 3D Gaussian fittings of various 3D objects. Our experimental results show that text-to-image diffusion models can be effectively adapted for 3D content generation, bridging the gap between 2D and 3D modeling.

Survival clustering approaches estimate time-to-event prediction and clustering simultaneously. Current methods, however, are often limited by their reliance on single data modalities and their separation of risk prediction from patient subtyping. This creates a pressing need for a unified framework that can simultaneously discover patient subgroups and predict their conversion risk by integrating diverse biomarker data. Here, we introduce Multi-Modal Deep Clustering Survival Machines (MMDCSM), a unified framework that (1) encodes each modality via modality-specific MultiLayer Perceptrons (MLPs), (2) fuses embeddings into a joint representation, and (3) models survival outcomes through a mixture of Weibull expert distributions whose soft assignments simultaneously define patient subtypes and individualized survival curves. Applied to a cohort of 382 Mild Cognitive Impairment (MCI) patients, MMDCSM significantly outperformed existing methods in identifying distinct low- and high-risk subgroups for Alzheimer's disease (AD) conversion while delivering competitive accuracy in predicting each patient's personalized timeline to progression. Our model also identified key brain regions, including the hippocampus, that were most influential in distinguishing high-risk converters. This approach enables more accurate, early-stage risk stratification, paving the way for targeted interventions designed to delay or prevent disease progression.

Short-form social-media videos offer a rich, low-cost source of behavioural data, yet their weak or noisy labels and privacy constraints complicate their direct use in action recognition. We introduce an end-to-end pipeline, Hashtag2Action (H2A), that turns raw clips from short-form video platforms (i.e. TikTok) into a large-scale, ethically curated H2A dataset. The pipeline combines adaptive hashtag mining, metadata filtering, and vision-based frame validation to assemble 283,582 clips spanning 386 action categories with minimal manual effort. Using this collection, we pre-train a VideoMAE V2 backbone in a self-supervised manner and fine-tune it on UCF101, HMDB51, Kinetics-400, and Something-Something V2. With only 20% of the original VideoMAE V2 pre-training data, the model achieves 99.1% (UCF101), 86.1% (HMDB51), 85.5% (Kinetics-400), and 74.3% (SSv2) top-1 accuracy. These results show that carefully curated, weakly labelled short-form videos can support competitive downstream performance without additional annotation. To support reproducible research in short-form video understanding, the pre-trained and fine-tuning weights and metadata are publicly available at https://doi.org/10.57967/hf/3179.

In this paper, we introduce Latest Object Memory (LOM), a system for robustly tracking and continuously updating the latest states of objects by explicitly modeling their presence across video frames. LOM enables consistent tracking and accurate identity management across frames, enhancing both performance and reliability through the video segmentation process. Building upon LOM, we present Latest Object Memory Management (LOMM) for temporally consistent video instance segmentation, significantly improving long-term instance tracking. This enables consistent tracking and accurate identity management across frames, enhancing both performance and reliability through the video segmentation process. Moreover, we introduce Decoupled Object Association (DOA), a strategy that separately handles newly appearing and already existing objects. By leveraging our memory system, DOA accurately assigns object indices, improving matching accuracy and ensuring stable identity consistency, even in dynamic scenes where objects frequently appear and disappear. Extensive experiments and ablation studies demonstrate the superiority of our method over traditional approaches, setting a new state-of-the-art in video instance segmentation. Notably, our LOMM achieves an AP score of 54.0 on YouTube-VIS 2022, a dataset known for its challenging long videos. Project page: this https URL.

Vision foundation models and generative AI have recently made remarkable advances, enabling groundbreaking applications across computer vision, multimodal reasoning, and interactive AI systems. These models hold immense potential to enhance accessibility by powering assistive technologies that improve the daily lives of individuals with disabilities. At the same time, applying foundation models and generative AI in accessibility contexts presents key challenges, including fairness and bias, data scarcity, interpretability, and real-world deployment. This CV4A11y workshop brings together researchers from computer vision, human-computer interaction, and accessibility communities to identify challenges, explore emerging opportunities, and foster collaboration toward building inclusive, efficient, and responsible AI-powered accessibility systems.

Anomaly detection and segmentation pose an important task across sectors ranging from medical imaging analysis to industry quality control. However, current unsupervised approaches require training data to not contain any anomalies, a requirement that can be especially challenging in many medical imaging scenarios. In this paper, we propose Iterative Latent Token Masking, a self-supervised framework derived from a robust statistics point of view, translating an iterative model fitting with M-estimators to the task of anomaly detection. In doing so, this allows the training of unsupervised methods on datasets heavily contaminated with anomalous images. Our method stems from prior work on using Transformers, combined with a Vector Quantized-Variational Autoencoder, for anomaly detection, a method with state-of-the-art performance when trained on normal (non-anomalous) data. More importantly, we utilise the token masking capabilities of Transformers to filter out suspected anomalous tokens from each sample's sequence in the training set in an iterative self-supervised process, thus overcoming the difficulties of highly anomalous training data. Our work also highlights shortfalls in current state-of-the-art self-supervised, self-trained and unsupervised models when faced with small proportions of anomalous training data. We evaluate our method on whole-body PET data in addition to showing its wider application in more common computer vision tasks such as the industrial MVTec Dataset. Using varying levels of anomalous training data, our method showcases a superior performance over several state-of-the-art models, drawing attention to the potential of this approach.

This paper introduces a novel benchmark to study the impact and relationship of built environment elements on pedestrian collision prediction, intending to enhance environmental awareness in autonomous driving systems to prevent pedestrian injuries actively. We introduce a built environment detection task in large-scale panoramic images and a detection-based pedestrian collision frequency prediction task. We propose a baseline method that incorporates a collision prediction module into a state-of-the-art detection model to tackle both tasks simultaneously. Our experiments demonstrate a significant correlation between object detection of built environment elements and pedestrian collision frequency prediction. Our results are a stepping stone towards understanding the interdependencies between built environment conditions and pedestrian safety.

Curriculum design is a fundamental component of education. For example, when we learn mathematics at school, we build upon our knowledge of addition to learn multiplication. These and other concepts must be mastered before our first algebra lesson, which also reinforces our addition and multiplication skills. Designing a curriculum for teaching either a human or a machine shares the underlying goal of maximizing knowledge transfer from earlier to later tasks, while also minimizing forgetting of learned tasks. Prior research on curriculum design for image classification focuses on the ordering of training examples during a single offline task. Here, we investigate the effect of the order in which multiple distinct tasks are learned in a sequence. We focus on the online class-incremental continual learning setting, where algorithms or humans must learn image classes one at a time during a single pass through a dataset. We find that curriculum consistently influences learning outcomes for humans and for multiple continual machine learning algorithms across several benchmark datasets. We introduce a novel-object recognition dataset for human curriculum learning experiments and observe that curricula that are effective for humans are highly correlated with those that are effective for machines. As an initial step towards automated curriculum design for online class-incremental learning, we propose a novel algorithm, dubbed Curriculum Designer (CD), that designs and ranks curricula based on inter-class feature similarities. We find significant overlap between curricula that are empirically highly effective and those that are highly ranked by our CD. Our study establishes a framework for further research on teaching humans and machines to learn continuously using optimized curricula. Our code and data are available through this link.

Accurate medical image segmentation is of utmost importance for enabling automated clinical decision procedures. However, prevailing supervised deep learning approaches for medical image segmentation encounter significant challenges due to their heavy dependence on extensive labeled training data. To tackle this issue, we propose a novel self-supervised algorithm, $S^3-\mathrm{Net}$, which integrates a robust framework based on the proposed Inception Large Kernel Attention (I-LKA) modules. This architectural enhancement makes it possible to comprehensively capture contextual information while preserving local intricacies, thereby enabling precise semantic segmentation. Furthermore, considering that lesions in medical images often exhibit deformations, we leverage deformable convolution as an integral component to effectively capture and delineate lesion deformations for superior object boundary definition. Additionally, our self-supervised strategy emphasizes the acquisition of invariance to affine transformations, which is commonly encountered in medical scenarios. This emphasis on robustness with respect to geometric distortions significantly enhances the model's ability to accurately model and handle such distortions. To enforce spatial consistency and promote the grouping of spatially connected image pixels with similar feature representations, we introduce a spatial consistency loss term. This aids the network in effectively capturing the relationships among neighboring pixels and enhancing the overall segmentation quality. The $S^3-Net$ approach iteratively learns pixel-level feature representations for image content clustering in an end-to-end manner. Our experimental results on skin lesion and lung organ segmentation tasks show the superior performance of our method compared to the SOTA approaches.

Age-related Macular Degeneration (AMD) is a degenerative eye disease that causes central vision loss. Optical Coherence Tomography Angiography (OCTA) is an emerging imaging modality that aids in the diagnosis of AMD by displaying the pathogenic vessels in the subretinal space. In this paper, we investigate the effectiveness of OCTA from the view of deep classifiers. To the best of our knowledge, this is the first study that solely uses OCTA for AMD stage grading. By developing a 2D classifier based on OCTA projections, we identify that segmentation errors in retinal layers significantly affect the accuracy of classification. To address this issue, we propose analyzing 3D OCTA volumes directly using a 2D convolutional neural network trained with additional projection supervision. Our experimental results show that we achieve over 80% accuracy on a four-stage grading task on both error-free and error-prone test sets, which is significantly higher than 60%, the accuracy of human experts. This demonstrates that OCTA provides sufficient information for AMD stage grading and the proposed 3D volume analyzer is more robust when dealing with OCTA data with segmentation errors.