ANCHOR: Integrating Adversarial Training with Hard-mined Supervised Contrastive Learning for Robust Representation Learning

Neural networks have changed the way machines interpret the world. At their core, they learn by following gradients, adjusting their parameters step by step until they identify the most discriminant patterns in the data. This process gives them their strength, yet it also opens the door to a hidden flaw. The very gradients that help a model learn can also be used to produce small, imperceptible tweaks that cause the model to completely alter its decision. Such tweaks are called adversarial attacks. These attacks exploit this vulnerability by adding tiny, imperceptible changes to images that, while leaving them identical to the human eye, cause the model to make wrong predictions. In this work, we propose Adversarially-trained Contrastive Hard-mining for Optimized Robustness (ANCHOR), a framework that leverages the power of supervised contrastive learning with explicit hard positive mining to enable the model to learn representations for images such that the embeddings for the images, their augmentations, and their perturbed versions cluster together in the embedding space along with those for other images of the same class while being separated from images of other classes. This alignment helps the model focus on stable, meaningful patterns rather than fragile gradient cues. On CIFAR-10, our approach achieves impressive results for both clean and robust accuracy under PGD-20 (epsilon = 0.031), outperforming standard adversarial training methods. Our results indicate that combining adversarial guidance with hard-mined contrastive supervision helps models learn more structured and robust representations, narrowing the gap between accuracy and robustness.

Open Multimodal Retrieval-Augmented Factual Image Generation

Large Multimodal Models (LMMs) have achieved remarkable progress in generating photorealistic and prompt-aligned images, but they often produce outputs that contradict verifiable knowledge, especially when prompts involve fine-grained attributes or time-sensitive events. Conventional retrieval-augmented approaches attempt to address this issue by introducing external information, yet they are fundamentally incapable of grounding generation in accurate and evolving knowledge due to their reliance on static sources and shallow evidence integration. To bridge this gap, we introduce ORIG, an agentic open multimodal retrieval-augmented framework for Factual Image Generation (FIG), a new task that requires both visual realism and factual grounding. ORIG iteratively retrieves and filters multimodal evidence from the web and incrementally integrates the refined knowledge into enriched prompts to guide generation. To support systematic evaluation, we build FIG-Eval, a benchmark spanning ten categories across perceptual, compositional, and temporal dimensions. Experiments demonstrate that ORIG substantially improves factual consistency and overall image quality over strong baselines, highlighting the potential of open multimodal retrieval for factual image generation.

CRAG-MM: Multi-modal Multi-turn Comprehensive RAG Benchmark

Wearable devices such as smart glasses are transforming the way people interact with their surroundings, enabling users to seek information regarding entities in their view. Multi-Modal Retrieval-Augmented Generation (MM-RAG) plays a key role in supporting such questions, yet there is still no comprehensive benchmark for this task, especially regarding wearables scenarios. To fill this gap, we present CRAG-MM -- a Comprehensive RAG benchmark for Multi-modal Multi-turn conversations. CRAG-MM contains a diverse set of 6.5K (image, question, answer) triplets and 2K visual-based multi-turn conversations across 13 domains, including 6.2K egocentric images designed to mimic captures from wearable devices. We carefully constructed the questions to reflect real-world scenarios and challenges, including five types of image-quality issues, six question types, varying entity popularity, differing information dynamism, and different conversation turns. We design three tasks: single-source augmentation, multi-source augmentation, and multi-turn conversations -- each paired with an associated retrieval corpus and APIs for both image-KG retrieval and webpage retrieval. Our evaluation shows that straightforward RAG approaches achieve only 32% and 43% truthfulness on CRAG-MM single- and multi-turn QA, respectively, whereas state-of-the-art industry solutions have similar quality (32%/45%), underscoring ample room for improvement. The benchmark has hosted KDD Cup 2025, attracting about 1K participants and 5K submissions, with winning solutions improving baseline performance by 28%, highlighting its early impact on advancing the field.

Vision Transformer for Robust Occluded Person Reidentification in Complex Surveillance Scenes

Person re-identification (ReID) in surveillance is challenged by occlusion, viewpoint distortion, and poor image quality. Most existing methods rely on complex modules or perform well only on clear frontal images. We propose Sh-ViT (Shuffling Vision Transformer), a lightweight and robust model for occluded person ReID. Built on ViT-Base, Sh-ViT introduces three components: First, a Shuffle module in the final Transformer layer to break spatial correlations and enhance robustness to occlusion and blur; Second, scenario-adapted augmentation (geometric transforms, erasing, blur, and color adjustment) to simulate surveillance conditions; Third, DeiT-based knowledge distillation to improve learning with limited labels.To support real-world evaluation, we construct the MyTT dataset, containing over 10,000 pedestrians and 30,000+ images from base station inspections, with frequent equipment occlusion and camera variations. Experiments show that Sh-ViT achieves 83.2% Rank-1 and 80.1% mAP on MyTT, outperforming CNN and ViT baselines, and 94.6% Rank-1 and 87.5% mAP on Market1501, surpassing state-of-the-art methods.In summary, Sh-ViT improves robustness to occlusion and blur without external modules, offering a practical solution for surveillance-based personnel monitoring.

PSScreen V2: Partially Supervised Multiple Retinal Disease Screening

In this work, we propose PSScreen V2, a partially supervised self-training framework for multiple retinal disease screening. Unlike previous methods that rely on fully labelled or single-domain datasets, PSScreen V2 is designed to learn from multiple partially labelled datasets with different distributions, addressing both label absence and domain shift challenges. To this end, PSScreen V2 adopts a three-branch architecture with one teacher and two student networks. The teacher branch generates pseudo labels from weakly augmented images to address missing labels, while the two student branches introduce novel feature augmentation strategies: Low-Frequency Dropout (LF-Dropout), which enhances domain robustness by randomly discarding domain-related low-frequency components, and Low-Frequency Uncertainty (LF-Uncert), which estimates uncertain domain variability via adversarially learned Gaussian perturbations of low-frequency statistics. Extensive experiments on multiple in-domain and out-of-domain fundus datasets demonstrate that PSScreen V2 achieves state-of-the-art performance and superior domain generalization ability. Furthermore, compatibility tests with diverse backbones, including the vision foundation model DINOv2, as well as evaluations on chest X-ray datasets, highlight the universality and adaptability of the proposed framework. The codes are available at https://github.com/boyiZheng99/PSScreen_V2.

FlexICL: A Flexible Visual In-context Learning Framework for Elbow and Wrist Ultrasound Segmentation

Elbow and wrist fractures are the most common fractures in pediatric populations. Automatic segmentation of musculoskeletal structures in ultrasound (US) can improve diagnostic accuracy and treatment planning. Fractures appear as cortical defects but require expert interpretation. Deep learning (DL) can provide real-time feedback and highlight key structures, helping lightly trained users perform exams more confidently. However, pixel-wise expert annotations for training remain time-consuming and costly. To address this challenge, we propose FlexICL, a novel and flexible in-context learning (ICL) framework for segmenting bony regions in US images. We apply it to an intra-video segmentation setting, where experts annotate only a small subset of frames, and the model segments unseen frames. We systematically investigate various image concatenation techniques and training strategies for visual ICL and introduce novel concatenation methods that significantly enhance model performance with limited labeled data. By integrating multiple augmentation strategies, FlexICL achieves robust segmentation performance across four wrist and elbow US datasets while requiring only 5% of the training images. It outperforms state-of-the-art visual ICL models like Painter, MAE-VQGAN, and conventional segmentation models like U-Net and TransUNet by 1-27% Dice coefficient on 1,252 US sweeps. These initial results highlight the potential of FlexICL as an efficient and scalable solution for US image segmentation well suited for medical imaging use cases where labeled data is scarce.

AesCrop: Aesthetic-driven Cropping Guided by Composition

Aesthetic-driven image cropping is crucial for applications like view recommendation and thumbnail generation, where visual appeal significantly impacts user engagement. A key factor in visual appeal is composition--the deliberate arrangement of elements within an image. Some methods have successfully incorporated compositional knowledge through evaluation-based and regression-based paradigms. However, evaluation-based methods lack globality while regression-based methods lack diversity. Recently, hybrid approaches that integrate both paradigms have emerged, bridging the gap between these two to achieve better diversity and globality. Notably, existing hybrid methods do not incorporate photographic composition guidance, a key attribute that defines photographic aesthetics. In this work, we introduce AesCrop, a composition-aware hybrid image-cropping model that integrates a VMamba image encoder, augmented with a novel Mamba Composition Attention Bias (MCAB) and a transformer decoder to perform end-to-end rank-based image cropping, generating multiple crops along with the corresponding quality scores. By explicitly encoding compositional cues into the attention mechanism, MCAB directs AesCrop to focus on the most compositionally salient regions. Extensive experiments demonstrate that AesCrop outperforms current state-of-the-art methods, delivering superior quantitative metrics and qualitatively more pleasing crops.

Retrieval-Augmented Search for Large-Scale Map Collections with ColPali

Multimodal approaches have shown great promise for searching and navigating digital collections held by libraries, archives, and museums. In this paper, we introduce map-RAS: a retrieval-augmented search system for historic maps. In addition to introducing our framework, we detail our publicly-hosted demo for searching 101,233 map images held by the Library of Congress. With our system, users can multimodally query the map collection via ColPali, summarize search results using Llama 3.2, and upload their own collections to perform inter-collection search. We articulate potential use cases for archivists, curators, and end-users, as well as future work with our system in both machine learning and the digital humanities. Our demo can be viewed at: http://www.mapras.com.

Do Students Debias Like Teachers? On the Distillability of Bias Mitigation Methods

Knowledge distillation (KD) is an effective method for model compression and transferring knowledge between models. However, its effect on model's robustness against spurious correlations that degrade performance on out-of-distribution data remains underexplored. This study investigates the effect of knowledge distillation on the transferability of ``debiasing'' capabilities from teacher models to student models on natural language inference (NLI) and image classification tasks. Through extensive experiments, we illustrate several key findings: (i) overall the debiasing capability of a model is undermined post-KD; (ii) training a debiased model does not benefit from injecting teacher knowledge; (iii) although the overall robustness of a model may remain stable post-distillation, significant variations can occur across different types of biases; and (iv) we pin-point the internal attention pattern and circuit that causes the distinct behavior post-KD. Given the above findings, we propose three effective solutions to improve the distillability of debiasing methods: developing high quality data for augmentation, implementing iterative knowledge distillation, and initializing student models with weights obtained from teacher models. To the best of our knowledge, this is the first study on the effect of KD on debiasing and its interenal mechanism at scale. Our findings provide understandings on how KD works and how to design better debiasing methods.

Foundation Models in Dermatopathology: Skin Tissue Classification

The rapid generation of whole-slide images (WSIs) in dermatopathology necessitates automated methods for efficient processing and accurate classification. This study evaluates the performance of two foundation models, UNI and Virchow2, as feature extractors for classifying WSIs into three diagnostic categories: melanocytic, basaloid, and squamous lesions. Patch-level embeddings were aggregated into slide-level features using a mean-aggregation strategy and subsequently used to train multiple machine learning classifiers, including logistic regression, gradient-boosted trees, and random forest models. Performance was assessed using precision, recall, true positive rate, false positive rate, and the area under the receiver operating characteristic curve (AUROC) on the test set. Results demonstrate that patch-level features extracted using Virchow2 outperformed those extracted via UNI across most slide-level classifiers, with logistic regression achieving the highest accuracy (90%) for Virchow2, though the difference was not statistically significant. The study also explored data augmentation techniques and image normalization to enhance model robustness and generalizability. The mean-aggregation approach provided reliable slide-level feature representations. All experimental results and metrics were tracked and visualized using WandB.ai, facilitating reproducibility and interpretability. This research highlights the potential of foundation models for automated WSI classification, providing a scalable and effective approach for dermatopathological diagnosis while paving the way for future advancements in slide-level representation learning.