weakly supervised localization

This work addresses the limitation of existing weakly supervised video anomaly detection methods, which primarily focus on temporal localization while lacking precise spatial awareness, thereby hindering interpretability in real-world applications. We propose a patch-based spatiotemporal anomaly localization framework that jointly models the temporal and spatial locations of anomalies using only video-level labels. Leveraging multiple instance learning, our approach infers region-level anomaly scores from grid-level patch features and introduces a novel neighborhood-aware Top-k spatiotemporal selection strategy to generate fine-grained spatial anomaly maps without requiring bounding box supervision. Additionally, we provide frame-level bounding box annotations for two widely used datasets. Extensive experiments demonstrate that our method significantly outperforms state-of-the-art approaches across multiple benchmarks, achieving substantial improvements in spatiotemporal localization accuracy. The code, models, and new annotations are publicly released.

To address the challenge of localizing tampered regions in real-world scenarios where pixel-level annotations are unavailable, this paper proposes a weakly supervised forgery localization framework. Methodologically, it fuses multi-view activation maps from an image-level detection network (WCBnet) with region priors derived from pre-trained segmentation models (e.g., Segment Anything, DeepLab, PSPNet), and refines the localization posterior distribution via Bayesian inference. Its key innovation lies in being the first to integrate multi-resolution feature learning with segmentation priors for weakly supervised localization—achieving high-precision, interpretable, and pixel-level localization without any pixel-level supervision. Evaluated on multiple benchmark datasets, the framework achieves state-of-the-art performance, significantly outperforming existing methods in localization accuracy. This demonstrates both the feasibility and practicality of reliable forgery region localization under purely image-level supervision.

This work addresses the high cost of pixel-level annotation in medical image segmentation by proposing a novel approach that integrates diagnostic summary statistics—such as lesion area—with extremely sparse pixel-level weak supervision. The method employs a multi-objective loss function that jointly optimizes image reconstruction quality, fidelity to the provided summary statistics, and foreground overlap constraints derived from the sparse annotations. This design effectively overcomes the inherent limitations of relying solely on aggregate statistics, which often fail to yield precise spatial delineation. Evaluated on diverse datasets including natural images, breast ultrasound, and renal CT tumor scans, the proposed framework significantly outperforms baselines using only summary statistics. To the best of our knowledge, this is the first method to successfully synergize global statistical cues with minimal pixel-wise supervision, achieving high segmentation accuracy while substantially reducing annotation burden.

To address the high cost of pixel-level annotations and low-quality pseudo-labels in weakly supervised OCT image segmentation, this paper proposes a dual-guided (structural and textual) pseudo-label generation framework. Methodologically: (1) a structure-aware layer enhancement module is designed to improve anatomical layer segmentation robustness; (2) a dual-path text-guided mechanism integrates image-level label-derived textual descriptions with synthetically generated descriptive texts to achieve vision–semantics cross-modal alignment; (3) the framework incorporates CLIP-driven cross-domain text embeddings, a dual-branch visual encoder, and an iterative pseudo-label refinement strategy. Evaluated on three public OCT datasets, the method achieves significant mIoU improvements over existing weakly supervised approaches, establishing new state-of-the-art performance. The source code and pretrained models are publicly released.

This work addresses the challenge of efficiently localizing an unknown number of anomalous regions in large-scale spatially dependent data. The authors propose SPLADE, a two-stage method that integrates intelligent sampling with boundary estimation to simultaneously and consistently estimate both the number and boundaries of multiple axis-aligned anomalous patches under general spatial dependence structures—without requiring full spatial grid segmentation. By leveraging a uniform Gaussian approximation and an efficient search strategy, SPLADE substantially improves computational efficiency and localization accuracy. Experimental results demonstrate that SPLADE outperforms existing approaches on both synthetic and real-world video surveillance datasets, achieving faster runtime, higher localization precision, and robustness to strong spatial dependencies.

This work addresses the high cost and expert dependency of fine-grained spatiotemporal annotation in laparoscopic videos, which hinders the development of surgical instrument segmentation models. The authors propose a human-in-the-loop weakly supervised learning framework that, for the first time, integrates temporally consistent class activation maps generated by foundation models with active learning. By iteratively refining dual-objective losses—comprising video-level classification and pixel-level segmentation losses—using only video-level weak labels and expert-corrected image-level masks, the method progressively generates high-quality pseudo-masks. This approach substantially reduces annotation burden, achieving comparable performance with 50% less manual labeling effort, and enables efficient scaling to large-scale clinical datasets.

This study addresses the challenges of sparse lesion annotations and high expert labeling costs in early diabetic retinopathy (DR) screening, which hinder the performance of deep learning models. To overcome these limitations, the authors propose the SAFE framework, a two-stage weakly supervised learning approach that integrates a dual-branch patch embedding network, contrastive learning, and multi-embedding space ensembling. By leveraging spatial-semantic proximity to propagate sparse annotations and incorporating an abstention mechanism to ensure label reliability, SAFE achieves a patch-level classification accuracy of 0.9886 between healthy and diseased regions. The method significantly improves downstream DR classification performance, attaining an F1 score and AUPRC of 0.545, and is validated by ophthalmologists to focus on clinically relevant lesions, demonstrating both high interpretability and practical utility.

This work addresses the challenge of object detection in medical imaging under weak supervision, where only point-level annotations are available, making accurate bounding box inference difficult amidst complex scenarios such as anatomical overlap and scale variation. To this end, the authors propose DExTeR, a Point-DETR-based Transformer architecture for point-to-box regression. DExTeR enhances category-specific feature extraction through class-guided deformable attention, improves instance discrimination via a CLICK-MoE (Category-Instance-Common Knowledge Mixture of Experts) module, and incorporates a multi-point consistency training strategy to boost robustness against variations in annotation placement. Evaluated on three diverse medical imaging datasets—endoscopy, chest X-ray, and endoscopic ultrasound—DExTeR achieves state-of-the-art detection performance while substantially reducing annotation costs.

This work addresses the challenge of high annotation costs and the availability of only video-level weak labels in video anomaly detection by proposing a novel weakly supervised approach that, for the first time, jointly models spatiotemporal local anomalies under such constraints. The method treats normal and anomalous videos as negative and positive bags, respectively, within a multiple instance learning (MIL) framework. It integrates spatiotemporal feature extraction with a classifier-driven anomaly scoring mechanism and introduces a multiple instance ranking loss to effectively leverage video-level labels for pixel-level anomaly localization. Experimental results on the UCF Crime2Local dataset demonstrate that the proposed method accurately detects localized spatiotemporal anomalies using only video-level supervision.