Vision-Language Models

Accurate and timely identification of construction hazards around workers is essential for preventing workplace accidents. While large vision-language models (VLMs) demonstrate strong contextual reasoning capabilities, their high computational requirements limit their applicability in near real-time construction hazard detection. In contrast, small vision-language models (sVLMs) with fewer than 4 billion parameters offer improved efficiency but often suffer from reduced accuracy and hallucination when analyzing complex construction scenes. To address this trade-off, this study proposes a detection-guided sVLM framework that integrates object detection with multimodal reasoning for contextual hazard identification. The framework first employs a YOLOv11n detector to localize workers and construction machinery within the scene. The detected entities are then embedded into structured prompts to guide the reasoning process of sVLMs, enabling spatially grounded hazard assessment. Within this framework, six sVLMs (Gemma-3 4B, Qwen-3-VL 2B/4B, InternVL-3 1B/2B, and SmolVLM-2B) were evaluated in zero-shot settings on a curated dataset of construction site images with hazard annotations and explanatory rationales. The proposed approach consistently improved hazard detection performance across all models. The best-performing model, Gemma-3 4B, achieved an F1-score of 50.6%, compared to 34.5% in the baseline configuration. Explanation quality also improved significantly, with BERTScore F1 increasing from 0.61 to 0.82. Despite incorporating object detection, the framework introduces minimal overhead, adding only 2.5 ms per image during inference. These results demonstrate that integrating lightweight object detection with small VLM reasoning provides an effective and efficient solution for context-aware construction safety hazard detection.

The increasing use of Online Vision Language Models (OVLMs) for processing images has introduced significant privacy risks, as individuals frequently upload images for various utilities, unaware of the potential for privacy violations. Images contain relationships that relate to Personally Identifiable Information (PII), where even seemingly harmless details can indirectly reveal sensitive information through surrounding clues. This paper explores the critical issue of PII disclosure in images uploaded to OVLMs and its implications for user privacy. We investigate how the extraction of contextual relationships from images can lead to direct (explicit) or indirect (implicit) exposure of PII, significantly compromising personal privacy. Furthermore, we propose methods to protect privacy while preserving the intended utility of the images in Vision Language Model (VLM)-based applications. Our evaluation demonstrates the efficacy of these techniques, highlighting the delicate balance between maintaining utility and protecting privacy in online image processing environments. Index Terms-Personally Identifiable Information (PII), Privacy, Utility, privacy concerns, sensitive information

It is critical for vision-language models (VLMs) to comprehensively understand visual, temporal, and textual cues. However, despite rapid progress in multimodal modeling, video understanding performance still lags behind text-based reasoning. In this work, we find that progress is even worse than previously assumed: commonly reported long video understanding benchmarks contain 40-60% of questions that can be answered using text cues alone. Furthermore, we find that these issues are also pervasive in widely used post-training datasets, potentially undercutting the ability of post-training to improve VLM video understanding performance. Guided by this observation, we introduce VidGround as a simple yet effective solution: using only the actual visually grounded questions without any linguistic biases for post-training. When used in tandem with RL-based post-training algorithms, this simple technique improves performance by up to 6.2 points relative to using the full dataset, while using only 69.1% of the original post-training data. Moreover, we show that data curation with a simple post-training algorithm outperforms several more complex post-training techniques, highlighting that data quality is a major bottleneck for improving video understanding in VLMs. These results underscore the importance of curating post-training data and evaluation benchmarks that truly require visual grounding to advance the development of more capable VLMs. Project page: http://vidground.etuagi.com.

Automated fact-checking is a crucial task that supports a responsible information ecosystem. While recent research has progressed from text-only to multimodal fact-checking, a prevailing assumption is that incorporating visual evidence universally improves verification accuracy. In this work, we challenge this assumption and show that the indiscriminate use of visual evidence can reduce accuracy. Building on this finding, we propose AMuFC, a modular fact-checking framework that employs two collaborative vision-language models with distinct roles to enable the adaptive use of visual evidence. Experimental results on three datasets, including WebFC, introduced in this study, demonstrate the effectiveness of adaptive visual evidence use in fact-checking.

The rapid advancement of AI generated content (AIGC) has blurred the boundaries between real and synthetic images, exposing the limitations of existing deepfake detectors that often overfit to specific generative models. This adaptability crisis calls for a fundamental reexamination of the intrinsic physical characteristics that distinguish natural from AI-generated images. In this paper, we address two critical research questions: (1) What physical features can stably and robustly discriminate AI generated images across diverse datasets and generative architectures? (2) Can these objective pixel-level features be integrated into multimodal models like CLIP to enhance detection performance while mitigating the unreliability of language-based information? To answer these questions, we conduct a comprehensive exploration of 15 physical features across more than 20 datasets generated by various GANs and diffusion models. We propose a novel feature selection algorithm that identifies five core physical features including Laplacian variance, Sobel statistics, and residual noise variance that exhibit consistent discriminative power across all tested datasets. These features are then converted into text encoded values and integrated with semantic captions to guide image text representation learning in CLIP. Extensive experiments demonstrate that our method achieves state-of-the-art performance on multiple Genimage benchmarks, with near-perfect accuracy (99.8%) on datasets such as Wukong and SDv1.4. By bridging pixel level authenticity with semantic understanding, this work pioneers the use of physically grounded features for trustworthy vision language modeling and opens new directions for mitigating hallucinations and textual inaccuracies in large multimodal models.

Reasoning VLMs can become more accurate while progressively losing visual grounding as they think. This creates task-conditional danger zones where low-entropy predictions are confident but ungrounded, a failure mode text-only monitoring cannot detect. Evaluating three reasoning VLMs on MathVista, HallusionBench, and MMMU_Pro, we find a pervasive evidence-collapse phenomenon: attention to annotated evidence regions drops substantially, often losing over half of evidence mass, as reasoning unfolds. Full-response entropy is the most reliable text-only uncertainty signal under cross-dataset transfer, yet adding vision features with a single global linear rule is brittle and often degrades transfer. An entropy-vision interaction model reveals a task-conditional regime: lowentropy, visually disengaged predictions are hazardous on sustained visual-reference tasks but benign on symbolic tasks. Using this structure, a targeted vision veto reduces selective risk by up to 1.9 percentage points at 90% coverage, while avoiding degradations where disengagement is expected. The results support task-aware multimodal monitoring for safe deployment under distribution shift.

On-screen learning behavior provides valuable insights into how students seek, use, and create information during learning. Analyzing on-screen behavioral engagement is essential for capturing students' cognitive and collaborative processes. The recent development of Vision Language Models (VLMs) offers new opportunities to automate the labor-intensive manual coding often required for multimodal video data analysis. In this study, we compared the performance of both leading closed-source VLMs (Claude-3.7-Sonnet, GPT-4.1) and open-source VLM (Qwen2.5-VL-72B) in single- and multi-agent settings for automated coding of screen recordings in collaborative learning contexts based on the ICAP framework. In particular, we proposed and compared two multi-agent frameworks: 1) a three-agent workflow multi-agent system (MAS) that segments screen videos by scene and detects on-screen behaviors using cursor-informed VLM prompting with evidence-based verification; 2) an autonomous-decision MAS inspired by ReAct that iteratively interleaves reasoning, tool-like operations (segmentation/ classification/ validation), and observation-driven self-correction to produce interpretable on-screen behavior labels. Experimental results demonstrated that the two proposed MAS frameworks achieved viable performance, outperforming the single VLMs in scene and action detection tasks. It is worth noting that the workflow-based agent achieved best on scene detection, and the autonomous-decision MAS achieved best on action detection. This study demonstrates the effectiveness of VLM-based Multi-agent System for video analysis and contributes a scalable framework for multimodal data analytics.

Recent vision-language models (VLMs) typically rely on a single vision encoder trained with contrastive image-text objectives, such as CLIP-style pretraining. While contrastive encoders are effective for cross-modal alignment and retrieval, self-supervised visual encoders often capture richer dense semantics and exhibit stronger robustness on recognition and understanding tasks. In this work, we investigate how to scale the fusion of these complementary visual representations for vision-language modeling. We propose CoME-VL: Complementary Multi-Encoder Vision-Language, a modular fusion framework that integrates a contrastively trained vision encoder with a self-supervised DINO encoder. Our approach performs representation-level fusion by (i) entropy-guided multi-layer aggregation with orthogonality-constrained projections to reduce redundancy, and (ii) RoPE-enhanced cross-attention to align heterogeneous token grids and produce compact fused visual tokens. The fused tokens can be injected into a decoder-only LLM with minimal changes to standard VLM pipelines. Extensive experiments across diverse vision-language benchmarks demonstrate that CoME-VL consistently outperforms single-encoder baselines. In particular, we observe an average improvement of 4.9% on visual understanding tasks and 5.4% on grounding tasks. Our method achieves state-of-the-art performance on RefCOCO for detection while improving over the baseline by a large margin. Finally, we conduct ablation studies on layer merging, non-redundant feature mixing, and fusion capacity to evaluate how complementary contrastive and self-supervised signals affect VLM performance.

Predicting product quality from multimodal item information is critical in cold-start scenarios, where user interaction history is unavailable and predictions must rely on images and textual metadata. However, existing vision-language models typically depend on large architectures and/or extensive external datasets, resulting in high computational cost. To address this, we propose EffiMiniVLM, a compact dual-encoder vision-language regression framework that integrates an EfficientNet-B0 image encoder and a MiniLM-based text encoder with a lightweight regression head. To improve training sample efficiency, we introduce a weighted Huber loss that leverages rating counts to emphasize more reliable samples, yielding consistent performance gains. Trained using only 20% of the Amazon Reviews 2023 dataset, the proposed model contains 27.7M parameters and requires 6.8 GFLOPs, yet achieves a CES score of 0.40 with the lowest resource cost in the benchmark. Despite its small size, it remains competitive with significantly larger models, achieving comparable performance while being approximately 4x to 8x more resource-efficient than other top-5 methods and being the only approach that does not use external datasets. Further analysis shows that scaling the data to 40% alone allows our model to overtake other methods, which use larger models and datasets, highlighting strong scalability despite the model's compact design.

Vision-Language Models (VLMs) have advanced rapidly within the unified Transformer architecture, yet their deployment on resource-constrained devices remains challenging due to high computational complexity. While pruning has emerged as an effective technique for compressing VLMs, existing approaches predominantly focus on a single mode by pruning either parameters or tokens, neglecting fully exploring the inherent redundancy in each mode, which leads to substantial performance degradation at high pruning ratios. To address the above limitations, we propose Collaborative Multi-Mode Pruning (CoMP), a novel framework tailored for VLMs by performing joint parameter and token pruning. Specifically, we first design a Collaborative Importance Metric (CIM) that investigates the mutual interference between the coupled parameters and tokens. It incorporates distinct significance of tokens into the computation of parameter importance scores, while simultaneously mitigating the affect of pruned parameters on token importance scores. Moreover, we develop a Multi-Mode Pruning Strategy (MPS) that decomposes the overall pruning process into a sequence of pruning stages, while in each stage we estimate the priory of different pruning modes based on their pruning cost and adaptively shift to the optimal one. Additionally, MPS integrates the historical cost and random exploration, in order to achieve a stable pruning process and avoid local optimum. Extensive experiments across various vision-language tasks and models demonstrate that our method effectively promotes the performance under high pruning ratios by comparing to the state-of-the-art approaches. The source code is available at https://github.com/Wuzimeng/CoMP.git.

As vision-language models (VLMs) are increasingly deployed in clinical decision support, more than accuracy is required: knowing when to trust their predictions is equally critical. Yet, a comprehensive and systematic investigation into the overconfidence of these models remains notably scarce in the medical domain. We address this gap through a comprehensive empirical study of confidence calibration in VLMs, spanning three model families (Qwen3-VL, InternVL3, LLaVA-NeXT), three model scales (2B--38B), and multiple confidence estimation prompting strategies, across three medical visual question answering (VQA) benchmarks. Our study yields three key findings: First, overconfidence persists across model families and is not resolved by scaling or prompting, such as chain-of-thought and verbalized confidence variants. Second, simple post-hoc calibration approaches, such as Platt scaling, reduce calibration error and consistently outperform the prompt-based strategy. Third, due to their (strict) monotonicity, these post-hoc calibration methods are inherently limited in improving the discriminative quality of predictions, leaving AUROC at the same level. Motivated by these findings, we investigate hallucination-aware calibration (HAC), which incorporates vision-grounded hallucination detection signals as complementary inputs to refine confidence estimates. We find that leveraging these hallucination signals improves both calibration and AUROC, with the largest gains on open-ended questions. On closed-ended questions, format-specific signals such as the logit margin show more informative. Overall, our findings suggest post-hoc calibration as standard practice for medical VLM deployment over raw confidence estimates, and highlight the practical usefulness of hallucination signals to enable more reliable use of VLMs in medical VQA.

Developing vision-language models (VLMs) that generalize across diverse tasks requires large-scale training datasets with diverse content. In English, such datasets are typically constructed by aggregating and curating numerous existing visual question answering (VQA) resources. However, this strategy does not readily extend to other languages, where VQA datasets remain limited in both scale and domain coverage. In this work, we introduce Jagle, the largest Japanese multimodal post-training dataset to date, comprising approximately 9.2 million instances across diverse tasks. Rather than relying on existing VQA datasets, we collect heterogeneous source data, including images, image-text pairs, and PDF documents, and generate VQA pairs through multiple strategies such as VLM-based QA generation, translation, and text rendering. Experiments demonstrate that a 2.2B model trained with Jagle surpasses InternVL3.5-2B in average score across ten Japanese evaluation tasks and comes within five points of Qwen3-VL-2B-Instruct. Moreover, combining Jagle with FineVision improves English performance over training with FineVision alone. These benefits generalize across model backbones and scales: LLM-jp-4-VL 9B beta, a 9B-scale model built on the Japanese-centric LLM-jp-4-8B-instruct, achieves Japanese performance comparable to Qwen3-VL-8B-Instruct. To facilitate reproducibility and future research, we release the dataset, models, and code.

Large vision-language models (VLMs) excel on general benchmarks but often lack robustness in medical imaging, where heterogeneous supervision induces cross-dataset interference and sensitivity to data regime (i.e., how the supervisory signals are mixed). In realistic clinical workflows, data and tasks arrive sequentially, so naive continual training further leads to catastrophic forgetting. To address these challenges, we propose MedQwen, a parameter-efficient medical VLM that couples a spectrally routed Mixture-of-Experts (MoE) with a theoretically grounded scaling rule that aligns low-rank updates with a full-rank, fully fine-tuned MoE, without changing the base architecture. Concretely, we initialize each expert from non-overlapping singular value decomposition (SVD) segments of the pretrained weight and introduce a residual compensation and scaling scheme to enable stable expert specialization and consistent routing under distribution shift. Across 23 medical datasets covering visual question answering, report generation, radiology classification, and hallucination mitigation, MedQwen achieves strong, reliable performance: it approaches full fine-tuning on zero-shot classification with 339$\times$ fewer trainable parameters, and reduces sequential forgetting to $\sim$5% where strong baselines degrade by $>$20-50%.

Vision-language models (VLMs) are vulnerable to adversarial image perturbations. Existing works based on adversarial training against task-specific adversarial examples are computationally expensive and often fail to generalize to unseen attack types. To address these limitations, we introduce Paraphrase-Decomposition-Aggregation (PDA), a training-free defense framework that leverages text augmentation to enhance VLM robustness under diverse adversarial image attacks. PDA performs prompt paraphrasing, question decomposition, and consistency aggregation entirely at test time, thus requiring no modification on the underlying models. To balance robustness and efficiency, we instantiate PDA as invariants that reduce the inference cost while retaining most of its robustness gains. Experiments on multiple VLM architectures and benchmarks for visual question answering, classification, and captioning show that PDA achieves consistent robustness gains against various adversarial perturbations while maintaining competitive clean accuracy, establishing a generic, strong and practical defense framework for VLMs during inference.

The rapid progress of large Vision-Language Models (VLMs) has enabled a wide range of applications, such as image understanding and Visual Question Answering (VQA). Query images are often uploaded to the cloud, where VLMs are typically hosted, hence efficient image compression becomes crucial. However, traditional human-centric codecs are suboptimal in this setting because they preserve many task-irrelevant details. Existing Image Coding for Machines (ICM) methods also fall short, as they assume a fixed set of downstream tasks and cannot adapt to prompt-driven VLMs with an open-ended variety of objectives. We propose a lightweight, plug-and-play, prompt-guided prefiltering module to identify image regions most relevant to the text prompt, and consequently to the downstream task. The module preserves important details while smoothing out less relevant areas to improve compression efficiency. It is codec-agnostic and can be applied before conventional and learned encoders. Experiments on several VQA benchmarks show that our approach achieves a 25-50% average bitrate reduction while maintaining the same task accuracy. Our source code is available at https://github.com/bardia-az/pgp-vlm-compression.

Optical Character Recognition (OCR) is increasingly regarded as a foundational capability for modern vision-language models (VLMs), enabling them not only to read text in images but also to support downstream reasoning in real-world visual question answering (VQA). However, practical applications further require reliable text anchors, i.e., accurately grounding queried text to its corresponding spatial region. To systematically evaluate this capability, we introduce TextAnchor-Bench (TABench), a benchmark for fine-grained text-region grounding, which reveals that both general-purpose and OCR-specific VLMs still struggle to establish accurate and stable text anchors. To address this limitation, we propose Q-Mask, a precise OCR framework built upon a causal query-driven mask decoder (CQMD). Inspired by chain-of-thought reasoning, Q-Mask performs causal visual decoding that sequentially generates query-conditioned visual masks before producing the final OCR output. This visual CoT paradigm disentangles where the text is from what the text is, enforcing grounded evidence acquisition prior to recognition and enabling explicit text anchor construction during inference. To train CQMD, we construct TextAnchor-26M, a large-scale dataset of image-text pairs annotated with fine-grained masks corresponding to specific textual elements, encouraging stable text-region correspondences and injecting strong spatial priors into VLM training. Extensive experiments demonstrate that Q-Mask substantially improves text anchoring and understanding across diverse visual scenes.

Public spaces such as transport hubs, city centres, and event venues require timely and reliable detection of potentially violent behaviour to support public safety. While automated video analysis has made significant progress, practical deployment remains constrained by latency, privacy, and resource limitations, particularly under edge-computing conditions. This paper presents the design and demonstrator-based deployment of a hybrid edge-based action detection system that combines skeleton-based motion analysis with vision-language models for semantic scene interpretation. Skeleton-based processing enables continuous, privacy-aware monitoring with low computational overhead, while vision-language models provide contextual understanding and zero-shot reasoning capabilities for complex and previously unseen situations. Rather than proposing new recognition models, the contribution focuses on a system-level comparison of both paradigms under realistic edge constraints. The system is implemented on a GPU-enabled edge device and evaluated with respect to latency, resource usage, and operational trade-offs using a demonstrator-based setup. The results highlight the complementary strengths and limitations of motioncentric and semantic approaches and motivate a hybrid architecture that selectively augments fast skeletonbased detection with higher-level semantic reasoning. The presented system provides a practical foundation for privacy-aware, real-time video analysis in public safety applications.

This paper presents a Multilingual Vision Large Language Model, named M-MiniGPT4. Our model exhibits strong vision-language understanding (VLU) capabilities across 11 languages. We utilize a mixture of native multilingual and translated data to push the multilingual VLU performance of the MiniGPT4 architecture. In addition, we propose a multilingual alignment training stage that uses parallel text corpora to further enhance the multilingual capabilities of our model. M-MiniGPT4 achieves 36% accuracy on the multilingual MMMU benchmark, outperforming state-of-the-art models in the same weight class, including foundation models released after the majority of this work was completed. We open-source our models, code, and translated datasets to facilitate future research in low-resource and multilingual settings.

Large Vision-Language Models have demonstrated exceptional performance in multimodal reasoning and complex scene understanding. However, these models still face significant hallucination issues, where outputs contradict visual facts. Recent research on hallucination mitigation has focused on retraining methods and Contrastive Decoding (CD) methods. While both methods perform well, retraining methods require substantial training resources, and CD methods introduce dual inference overhead. These factors hinder their practical applicability. To address the above issue, we propose a framework for dynamically detecting hallucination representations and performing hallucination-eliminating edits on these representations. With minimal additional computational cost, we achieve state-of-the-art performance on existing benchmarks. Extensive experiments demonstrate the effectiveness of our approach, highlighting its efficient and robust hallucination elimination capability and its powerful controllability over hallucinations. Code is available at https://github.com/ASGO-MM/HIRE

Training-free open-vocabulary remote sensing segmentation (OVRSS), empowered by vision-language models, has emerged as a promising paradigm for achieving category-agnostic semantic understanding in remote sensing imagery. Existing approaches mainly focus on enhancing feature representations or mitigating modality discrepancies to improve patch-level prediction accuracy. However, such independent prediction schemes are fundamentally misaligned with the intrinsic characteristics of remote sensing data. In real-world applications, remote sensing scenes are typically large-scale and exhibit strong spatial as well as semantic correlations, making isolated patch-wise predictions insufficient for accurate segmentation. To address this limitation, we propose ConInfer, a context-aware inference framework for OVRSS that performs joint prediction across multiple spatial units while explicitly modeling their inter-unit semantic dependencies. By incorporating global contextual cues, our method significantly enhances segmentation consistency, robustness, and generalization in complex remote sensing environments. Extensive experiments on multiple benchmark datasets demonstrate that our approach consistently surpasses state-of-the-art per-pixel VLM-based baselines such as SegEarth-OV, achieving average improvements of 2.80% and 6.13% on open-vocabulary semantic segmentation and object extraction tasks, respectively. The implementation code is available at: https://github.com/Dog-Yang/ConInfer

Visual document understanding typically requires separate retrieval and generation models, doubling memory and system complexity. We present Hydra, a dual-head approach that provides both ColBERT-style late-interaction retrieval and autoregressive generation from a single vision-language model. A single LoRA adapter, trained only for retrieval, is toggled at inference: enabling it produces multi-vector embeddings; disabling it recovers the base model's generation quality, with 426 of 426 language-model weight tensors byte-for-byte identical to a freshly-loaded Qwen3.5-4B. We identify two failure modes that can silently break generation in retrieval-fine-tuned VLMs (attention-mode restoration and lm_head preservation) plus an efficiency requirement (KV-cache-aware decoding); Hydra sidesteps the first two structurally and addresses the third in the decode loop. We release two scales, Hydra-4B and Hydra-0.8B, sharing LoRA hyperparameters (r=32, alpha=32) and optimisation recipe; data mix and projection dim differ across scales. The single-model design cuts peak GPU memory from 28.85 GB to 10.77 GB at 4B (62.7% reduction) and from 5.79 GB to 2.37 GB at 0.8B (59.1%) relative to a co-resident two-model deployment. A controlled ablation finds GritLM-style joint training matches Hydra's retrieval-only training on the evaluated modes while its LoRA-on generation mode collapses. A proof-of-concept on Qwen2.5-Omni-3B preserves generation equivalence on a non-Qwen3.5 backbone and transfers image retrieval within 2-8 pp of Hydra-4B, with zero-shot audio retrieval emerging through the frozen Whisper encoder.

Multimodal multiple-choice question answering (MCQA) provides a standardized and objectively measurable setting for evaluating vision-language models (VLMs). However, because the MCQA format incorporates the candidate choices into the input context, it introduces several unintended biases. Previous work has primarily focused on structural biases, such as preferences for certain choices. Instead, we argue that the choices act as textual priors, causing models to favor linguistically plausible options regardless of the visual content. We hypothesize and empirically verify that a model genuinely relies on visual evidence only when its multimodal distribution significantly diverges from its text-only distribution. Based on this observation we propose BUZZY,a training-free decoding method that corrects multimodal predictions by subtracting the text-only distribution. Experiments with five VLMs on five multimodal MCQA benchmarks demonstrate that BUZZY achieves the highest average accuracy among state-of-the-art methods while reducing inference latency by over 28% compared to prior contrastive decoding approaches. Overall, these results suggest that amplifying the visual signal by penalizing text-only preferences is key to efficient and robust multimodal MCQA reasoning. Code and additional resources are provided https://txxnrd.github.io/buzzy/.

Multimodal Large Language Models (MLLMs) perform strong vision-language reasoning under standard conditions but fail in extreme illumination, where RGB inputs lose irrevocable structure and semantics. We propose Event-MLLM, an event-enhanced model that performs all-light visual reasoning by dynamically fusing event streams with RGB frames. Two key components drive our approach: an Illumination Indicator - a learnable signal derived from a DINOv2 branch that represents exposure degradation and adaptively modulates event-RGB fusion - and an Illumination Correction Loss that aligns fused features with non-degraded (normal-light) semantics in the latent space, compensating for information lost in extreme lighting. We curate the first multi-illumination event-instruction corpus for MLLMs, with 2,241 event-RGB samples (around 6 QA pairs each) across diverse scenes and 17 brightness rates (0.05x - 20x), plus an instruct-following benchmark for reasoning, counting, and fine-grained recognition under extreme lighting. Experiments show that Event-MLLM markedly outperforms general-purpose, illumination-adaptive, and event-only baselines, setting a new state of the art in robust multimodal perception and reasoning under challenging illumination.

Long video understanding (LVU) remains a core challenge in multimodal learning. Although recent vision-language models (VLMs) have made notable progress, existing benchmarks mainly focus on either fine-grained perception or coarse summarization, offering limited insight into temporal understanding over long contexts. In this work, we define a scene as a coherent segment of a video in which both visual and semantic contexts remain consistent, aligning with human perception. This leads us to a key question: can current VLMs reason effectively over long, scene-level contexts? To answer this, we introduce a new benchmark, SceneBench, designed to provide scene-level challenges. Our evaluation reveals a sharp drop in accuracy when VLMs attempt to answer scene-level questions, indicating significant forgetting of long-range context. To further validate these findings, we propose Scene Retrieval-Augmented Generation (Scene-RAG), which constructs a dynamic scene memory by retrieving and integrating relevant context across scenes. This Scene-RAG improves VLM performance by +2.50%, confirming that current models still struggle with long-context retention. We hope SceneBench will encourage future research toward VLMs with more robust, human-like video comprehension.

Large Vision-Language Models (LVLMs) have achieved impressive performance across multimodal understanding and reasoning tasks, yet their internal safety mechanisms remain opaque and poorly controlled. In this work, we present a comprehensive framework for diagnosing and repairing unsafe channels within LVLMs (CARE). We first perform causal mediation analysis to identify neurons and layers that are causally responsible for unsafe behaviors. Based on these findings, we introduce a dual-modal safety subspace projection method that learns generalized safety subspaces for both visual and textual modalities through generalized eigen-decomposition between benign and malicious activations. During inference, activations are dynamically projected toward these safety subspaces via a hybrid fusion mechanism that adaptively balances visual and textual corrections, effectively suppressing unsafe features while preserving semantic fidelity. Extensive experiments on multiple safety benchmarks demonstrate that our causal-subspace repair framework significantly enhances safety robustness without degrading general multimodal capabilities, outperforming prior activation steering and alignment-based baselines. Additionally, our method exhibits good transferability, defending against unseen attacks.

Empowered by large-scale training, vision-language models (VLMs) achieve strong image and video understanding, yet their ability to perform spatial reasoning in both static scenes and dynamic videos remains limited. Recent advances try to handle this limitation by injecting geometry tokens from pretrained 3D foundation models into VLMs. Nevertheless, we observe that naive token fusion followed by standard fine-tuning in this line of work often leaves such geometric cues underutilized for spatial reasoning, as VLMs tend to rely heavily on 2D visual cues. In this paper, we propose GeoSR, a framework designed to make geometry matter by encouraging VLMs to actively reason with geometry tokens. GeoSR introduces two key components: (1) Geometry-Unleashing Masking, which strategically masks portions of 2D vision tokens during training to weaken non-geometric shortcuts and force the model to consult geometry tokens for spatial reasoning; and (2) Geometry-Guided Fusion, a gated routing mechanism that adaptively amplifies geometry token contributions in regions where geometric evidence is critical. Together, these designs unleash the potential of geometry tokens for spatial reasoning tasks. Extensive experiments on both static and dynamic spatial reasoning benchmarks demonstrate that GeoSR consistently outperforms prior methods and establishes new state-of-the-art performance by effectively leveraging geometric information. The project page is available at https://suhzhang.github.io/GeoSR/.

Autoregressive (AR) vision-language models (VLMs) have long dominated multimodal understanding, reasoning, and graphical user interface (GUI) grounding. Recently, discrete diffusion vision-language models (DVLMs) have shown strong performance in multimodal reasoning, offering bidirectional attention, parallel token generation, and iterative refinement. However, their potential for GUI grounding remains unexplored. In this work, we evaluate whether discrete DVLMs can serve as a viable alternative to AR models for GUI grounding. We adapt LLaDA-V for single-turn action and bounding-box prediction, framing the task as text generation from multimodal input. To better capture the hierarchical structure of bounding-box geometry, we propose a hybrid masking schedule that combines linear and deterministic masking, improving grounding accuracy by up to 6.1 points in Step Success Rate (SSR) over the GUI-adapted LLaDA-V trained with linear masking. Evaluations on four datasets spanning web, desktop, and mobile interfaces show that the adapted diffusion model with hybrid masking consistently outperforms the linear-masked variant and performs competitively with autoregressive counterparts despite limited pretraining. Systematic ablations reveal that increasing diffusion steps, generation length, and block length improves accuracy but also increases latency, with accuracy plateauing beyond a certain number of diffusion steps. Expanding the training data with diverse GUI domains further reduces latency by about 1.3 seconds and improves grounding accuracy by an average of 20 points across benchmarks. These results demonstrate that discrete DVLMs are a promising modeling framework for GUI grounding and represent an important step toward diffusion-based GUI agents.

Contrastive vision-language (V&L) models remain a popular choice for various applications. However, several limitations have emerged, most notably the limited ability of V&L models to learn compositional representations. Prior methods often addressed this limitation by generating custom training data to obtain hard negative samples. Hard negatives have been shown to improve performance on compositionality tasks, but are often specific to a single benchmark, do not generalize, and can cause substantial degradation of basic V&L capabilities such as zero-shot or retrieval performance, rendering them impractical. In this work we follow a different approach. We identify two root causes that limit compositionality performance of V&Ls: 1) Long training captions do not require a compositional representation; and 2) The final global pooling in the text and image encoders lead to a complete loss of the necessary information to learn binding in the first place. As a remedy, we propose two simple solutions: 1) We obtain short concept centric caption parts using standard NLP software and align those with the image; and 2) We introduce a parameter-free cross-modal attention-pooling to obtain concept centric visual embeddings from the image encoder. With these two changes and simple auxiliary contrastive losses, we obtain SOTA performance on standard compositionality benchmarks, while maintaining or improving strong zero-shot and retrieval capabilities. This is achieved without increasing inference cost. We release the code for this work at https://github.com/saic-fi/concept_centric_clip.

Contrastive Language-Image Pre-training (CLIP) demonstrates strong zero-shot generalization, but due to substantial computational and memory costs, distillation into lightweight models is required. Existing relational objectives do not explicitly model multidirectional relationships between teacher and student embeddings, potentially leaving the geometric relationships insufficiently constrained. This may disrupt the modality-gap structure important for zero-shot transfer. To address these limitations, we propose a relational distillation framework, CLIP-RD, which introduces two relational methods, Vertical Relational Distillation (VRD) and Cross Relational Distillation (XRD). VRD aligns teacher-student intra-modal similarity distributions to enforce consistent distillation strength across image and text embeddings. Meanwhile, XRD aligns the teacher-image-student-text and teacher-text-student-image similarity distributions to impose bidirectional cross-modal symmetry. By jointly modeling these multidirectional relational structures, CLIP-RD aligns the student's embedding geometry more faithfully to the teacher's, outperforming CLIP-KD by 1.8%p. This performance improvement is maintained across diverse architectures, teacher scales, retrieval tasks, downstream tasks, and corruption settings, with negligible additional training-time overhead.

Recent advances in multimodal vision-language models (VLMs) have enabled joint reasoning over visual and textual information, yet their application to planetary science remains largely unexplored. A key hindrance is the absence of large-scale datasets that pair real planetary imagery with detailed scientific descriptions. In this work, we introduce LLaVA-LE (Large Language-and-Vision Assistant for Lunar Exploration), a vision-language model specialized for lunar surface and subsurface characterization. To enable this capability, we curate a new large-scale multimodal lunar dataset, LUCID (LUnar Caption Image Dataset) consisting of 96k high-resolution panchromatic images paired with detailed captions describing lunar terrain characteristics, and 81k question-answer (QA) pairs derived from approximately 20k images in the LUCID dataset. Leveraging this dataset, we fine-tune LLaVA using a two-stage training curriculum: (1) concept alignment for domain-specific terrain description, and (2) instruction-tuned visual question answering. We further design evaluation benchmarks spanning multiple levels of reasoning complexity relevant to lunar terrain analysis. Evaluated against GPT and Gemini judges, LLaVA-LE achieves a 3.3x overall performance gain over Base LLaVA and 2.1x over our Stage 1 model, with a reasoning score of 1.070, exceeding the judge's own reference score, highlighting the effectiveness of domain-specific multimodal data and instruction tuning to advance VLMs in planetary exploration. Code is available at https://github.com/OSUPCVLab/LLaVA-LE.