Knowledge-Based Visual Question Answering

This paper presents our solution to the 2026 SoccerNet VQA Challenge. We first develop a cost-effective data synthesis pipeline driven by a Vision-Language Model (VLM), which systematically restructures raw domain data into diverse VQA samples, including concise answers and long-form responses. Second, we propose MSUE, a multi-expert question answering architecture that employs a Large Language Model (LLM) to dynamically dispatch questions to text, image, and video experts. These experts are instantiated as a strong text baseline Gemini3-Flash, a fine-tuned Qwen3-VL, and an external knowledge base, respectively, working collaboratively to enhance VQA performance. MSUE achieves an accuracy of \textbf{0.95} on the challenge benchmark, securing third place in the leaderboard.

Group Relative Policy Optimization (GRPO) and its variants, originally developed for Large Language Models (LLMs), have recently been applied to Multimodal LLMs and produced strong results. However, their coarse-grained holistic credit assignment from a single scalar advantage underfits vision-language (VL) tasks, where outputs are often long-form responses grounded in semantically rich images. To address this limitation, we exploit a structured signal that single-scalar formulations discard: the natural segmentation of long-form VL outputs. Concretely, we propose Segment-Decomposed GRPO (SD-GRPO), which z-normalizes verifiable per-segment rewards across the rollout group, yielding a vector of per-segment advantages in place of a single scalar. We evaluate SD-GRPO across three settings spanning controlled and real-world long-form VL generation, organized by increasing semantic entanglement across segments. On a controlled multi-panel dense-captioning task constructed from DOCCI, where segments are semantically independent, SD-GRPO consistently outperforms the GRPO baseline, with larger gains at higher segment counts. Extending to a controlled multi-chart long-form VQA task constructed from MultiChartQA, we show both theoretically and empirically that rollout-level rewards suffer from cross-segment credit misattribution that scales with output length. On a real-world scientific figure captioning task on the MMSci dataset, where subfigure captions share context across the figure, blending holistic and per-segment rewards further improves on both, suggesting per-segment normalization alone is insufficient when segments are semantically entangled. Finally, by integrating SD-GRPO into Dr. GRPO, we confirm that it can be applied to any GRPO framework with minimal implementation overhead to enhance long-form VL generation.

Visual Question Answering (VQA) is the task of answering questions about images, requiring the integration of multimodal input and reasoning. Modular approaches that incorporate logic-based representations into the reasoning component offer clear advantages over end-to-end trained systems, particularly in terms of interpretability. However, adapting or extending these representations when task requirements change can place a significant burden on developers. To address this challenge, we present an approach for distilling rules from Large Language Models (LLMs). Our method prompts an LLM to extend an initial VQA reasoning theory, expressed as an answer-set program, to meet new requirements of the task. Examples from VQA datasets guide the LLM, validate the results, and help correct erroneous rules by leveraging feedback from the ASP solver. We demonstrate that our approach is effective across diverse VQA datasets. Notably, only a few examples are needed to elicit correct rules from LLMs. Our experiments suggest that rule distillation from LLMs is a promising alternative to traditional data-driven rule learning approaches. Under consideration in Theory and Practice of Logic Programming (TPLP).

Augmenting large language models (LLMs) with retrieved web text has become a dominant paradigm, yet the web is not natively textual: existing systems depend on complex parsing pipelines that linearize HTML and discard layout, visual structure, and formatting. We introduce PixelRAG, a new retrieval-augmented method that represents websites in their native visual form and performs retrieval and reading entirely in pixel space, enabling an end-to-end architecture that eliminates text abstraction. PixelRAG is, to our knowledge, the first pipeline to operate over a full Wikipedia corpus in this form, scaling to a datastore of 30 million screenshot images with an efficient visual retrieval index. Built on an existing visual embedding model (i.e., Qwen3-VL-Embedding), PixelRAG further fine-tunes this model on screenshot data with carefully curated contrastive training data. Retrieved screenshots are then fed directly as pixel inputs to a VLM, without intermediate text conversion. PixelRAG consistently outperforms both no-retrieval and text-based RAG baselines, most surprisingly on widely studied text-centric tasks such as NQ and SimpleQA. It also achieves strong gains on multimodal open-domain QA (e.g., MMSearch), benchmarks over noisy news corpora (e.g., LiveVQA), and agentic benchmarks (e.g., MoNaCo), improving accuracy by up to 18.1% over text-based baselines. Finally, pixel representations enable a new efficiency lever for RAG through image compression, achieving up to 3x token cost reduction at lower resolutions while maintaining accuracy. Our results challenge the necessity of text representations in web retrieval, suggesting that web RAG can operate directly in the web's native visual form while improving both performance and efficiency.

Compositional visual question answering (VQA) represents a challenging yet fundamental task that requires models to comprehend novel combinations of previously learned concepts. The current methods often overlook the disentanglement of underlying concepts and are restricted in terms of their ability to effectively capture the compositional variation mechanism. Moreover, the state-of-the-art techniques depend on additional clues for training, which is not feasible in real-world VQA scenarios. To address these issues, in this paper, we introduce a novel Disentanglement-based EquivAriant Learning (DEAL) framework for compositional VQA, which is guided exclusively by ground-truth answers. In DEAL, we employ causality-inspired interventions to disentangle concepts derived from visual and textual inputs within a re-encoding framework. Based on the principle of equivariance, we subsequently perform a compositional transformation on the inference input and impose the equivariant constraint on the output to augment the compositional reasoning capacity of the model. Comprehensive experiments conducted on the benchmark CLEVR-CoGenT and GQA-SGL datasets validate the superiority of our proposed DEAL approach over the existing state-of-the-art methods for compositional VQA tasks in both visual and linguistic generalization settings.

Multimodal large language models (MLLMs) demonstrate remarkable visual understanding, yet their reliability in interactive settings is severely undermined by hallucination snowballing: a phenomenon where initial errors amplify across conversational turns, leading to a collapse in coherence. This failure reveals a fundamental vulnerability where models progressively neglect visual grounding in favor of over-relying on polluted textual history. Existing benchmarks are predominantly confined to single-turn VQA, which fail to capture the complex dynamics of error propagation in long-horizon interactions. To address this, we introduce MM-Snowball, the first benchmark for fine-grained diagnosis of hallucination snowballing within dialogues. Extensive evaluation shows that our benchmark poses a significant challenge even to advanced MLLMs and reveals the inefficacy of existing mitigation methods designed for single-turn VQA. To counteract this degradation, we propose Conflict-Aware Visual Rectification (CAVR). This training-free method mitigates snowballing through a synergistic dual-mechanism that refreshes visual grounding at the representation level and rectifies output distributions at the logit level, effectively re-anchoring the model to visual facts. Experiments demonstrate that CAVR achieves state-of-the-art performance, offering a promising path toward more reliable interactive AI. Data and code are available at: https://frenkie-chiang.github.io/MM-Snowball

Vision-language models (VLMs) can produce confident answers without relevant visual evidence, a failure mode known as mirage reasoning (Asadi et al., 2026). To that end, we study pre-release mirage detection: deciding whether a VLM answer should be released or withheld. Our model-agnostic method, Text-Conditioned Layer-wise Internal Alignment (TC-LIA), tracks question-image alignment across the layers of a frozen CLIP ViT-H/14 encoder, summarizing patch-text alignment by final similarity, late-layer top-k alignment, early-to-late gain, and slope. TC-LIA is purely unsupervised (fixed projections, fixed scoring weights, no labels, no training) and already delivers strong detection independently. Additionally, when combined with blank/noise detection, domain routing, and VLM self-assessment, it forms an ensemble whose supervised training improves performance but is an optional add-on. On 19,004 samples spanning ten VQA domains, fourteen state-of-the-art VLMs exhibit 57.3-75.0% base mirage rates. Our proposed TC-LIA alone cuts this to 7.5% with 83.5% Related/Unrelated/Blank-Noise classification accuracy, and the ensemble reaches 84.3-88.4% accuracy with 5.9-7.2% mirage rates (best joint result: 88.4% accuracy, 6.4% mirage rate). Notably, an ensemble trained on a single backbone transfers well to unseen backbones, with the best-transferring source staying within 1.2% accuracy points of per-backbone training across thirteen held-out VLMs.

As large-scale visual-document corpora such as arXiv papers and enterprise PDFs continue to grow, visual-document retrieval has gained increasing attention; yet it still lacks a deployable system that lexically indexes visual documents to serve queries without neural encoding at scale. Existing methods either achieve strong retrieval quality with VLM-based dense or multi-vector models but require neural query encoding at serving time, or avoid query encoding with OCR- or caption-based BM25 at the cost of time-consuming text extraction or generation. To fill this missing serving regime, we present V-SPLADE, an inference-free sparse retriever for visual-document retrieval. However, such inference-free multimodal learned sparse retrieval systems remain underexplored and have not yet shown dense-level effectiveness under high sparsity. We attribute this limitation to a lexical grounding problem: visual sparse representations often fail to capture the lexical content embedded in document images. To address this problem, we introduce caption-gated token supervision, a training-only signal that uses VLM-generated captions as lexical cues to activate retrieval-relevant vocabulary dimensions. With this supervision, V-SPLADE improves average NDCG@5 across six visual-document retrieval benchmarks by +13.8pp over the same-scale dense baseline and by up to +6.3pp over OCR- or caption-based BM25 baselines. On an 18.7M-document corpus, it more than doubles R@5 over the same-scale dense baseline and further improves competing retrievers through score fusion by up to +2.4pp R@5. Code will be released soon at https://github.com/naver/v-splade.

Vision-language models (VLMs) have achieved strong performance on visual question answering (VQA). To mitigate individual hallucinations and blind spots, aggregating diverse perspectives via multi-agent collaboration has emerged as a promising paradigm. While this approach has shown great success in textual QA, its potential in the multimodal domain remains under-explored. Existing multi-agent VQA methods predominantly adapt text-centric protocols, focusing on textual discussions while ignoring the alignment of visual information. In this work, we reveal a key insight: answer-level agreement is insufficient for reliable multi-agent VQA; \textit{aligned visual evidence} -- shared support from the image regions agents rely on -- is essential for trustworthy consensus. To leverage this insight, we propose EAGLE (\textbf{E}vidence-\textbf{A}ligned \textbf{G}rounded mu\textbf{L}ti-agent r\textbf{E}asoning), a training-free evidence-centered framework for coordinating multiple VLM agents. EAGLE explicitly exposes each agent's grounding regions as visual evidence, enables mutual verification over the evidence, and uses evidence consistency to guide final decision-making. Experiments on six VQA benchmarks show that EAGLE achieves best average performance across domains while remaining lightweight, interpretable, and practical for deployment.

Understanding long-form egocentric videos remains challenging for multimodal large language models (MLLMs) due to limited context length and insufficient grounding of fine-grained visual details. The recently proposed HD-EPIC benchmark highlights these limitations: even strong long-context models achieve relatively low performance across diverse video question answering tasks. In this paper, we propose a unified framework that decouples long-video reasoning into two complementary forms of evidence: semantic evidence and visual evidence. Semantic evidence captures global procedural structure through a coarse-to-fine extraction pipeline, while object-centric visual evidence preserves fine-grained grounding through bounding boxes and visual embeddings. During inference, we formulate reasoning as a query-conditioned evidence retrieval and integration process, dynamically selecting relevant information from both sources. Our approach achieves competitive performance in the HD-EPIC-VQA Challenge across multiple task categories. More broadly, our results demonstrate that explicitly structuring, retrieving, and integrating semantic and visual evidence is critical for effective long-video understanding with MLLMs.

Long-term memory is increasingly important for personalized AI agents, yet existing benchmarks and methods remain largely text-centric. Even when images are included, the user-specific information needed for later questions is typically recoverable from text alone, and most memory systems reduce image turns to generic captions. Yet images often carry personal information that text rarely states -- both explicit evidence, such as recurring user-associated entities, and implicit evidence, such as latent user facts inferred from visual or multimodal cues. We introduce a benchmark for personal visual memory that targets both forms of evidence, and propose VisualMem, a hybrid visual--text architecture that augments a text-memory backend with a structured personal visual memory module. Rather than collapsing images into captions, VisualMem uses conversational context to resolve identity, ownership, and durable user facts. Experiments show that VisualMem substantially outperforms prior memory systems on our benchmark while remaining competitive on standard text-memory benchmarks, indicating that personal visual memory is a distinct and important component of long-term memory for personalized AI agents.

Vision-Language-Action (VLA) models widely adopt pretrained Vision-Language Models (VLMs) as policy backbones, yet it remains unclear what kind of pretrained VLM representation is useful as a VLA initialization. In this paper, we study VLA initialization as a controlled representation-design problem along three axes: capability-level embodied VQA supervision, parameter-update strategy, and robot-data pretraining. Our experiments show that the original pretrained VLM representation is a key source of action performance. However, embodied VQA adaptation does not yield uniform gains: its benefit depends on downstream bottlenecks, and gains from different capability domains are not simply additive. For update strategy, LoRA provides a more reliable initialization than Full Finetune, indicating that overly reshaping the pretrained representation can weaken VLA initialization. Robot-data pretraining further improves VLA initialization, with the strongest variant obtained by staged LoRA-based training. Together, these findings suggest that effective VLM-to-VLA adaptation should inject action-relevant embodied and robot-trajectory signals while preserving the pretrained VLM representation that remains useful for action learning.

Reliable visual understanding in robot-assisted and minimally invasive surgery (RMIS/MIS) demands more than accurate masks: in clinical practice, clinicians pose language-like questions about procedural context, visibility, artefacts, and the presence of anatomical structures and surgical instruments, often under degraded views caused by occlusion, smoke, bleeding, and specular highlights. We present \textbf{RoboSurg-VQA}, a segmentation-aware visual question answering (VQA) benchmark built by repurposing public surgical segmentation datasets under a shared schema. Each frame is paired with a fixed set of clinically motivated questions spanning procedure context, anatomy (including region), imaging modality/view, surgical artefacts, image quality, and basic visibility and spatial attributes, with closed answer sets to enable consistent evaluation. To scale annotation, we generate candidate answers via constrained prompting with automatic validity and consistency checks, followed by human auditing to improve plausibility and label consistency. We report benchmark statistics, sanity baselines, and common evaluation challenges under challenging surgical conditions. The code will be available on https://github.com/ziyangwang007/Robosurg-VQA.

Continual Visual Question Answering (VQA) requires learning from non-stationary streams of visual inputs and questions while preserving past knowledge. Most prior methods adapt by updating a largely shared parameter set. This often leads to cross-level task interference, hindering accurate adaptation to the current task and object. To address this limitation, we propose HyLoVQA. It maintains a drift-resilient memory bank of anchors. The bank stores the content of visual objects and textual tasks, and they are updated using current input features. Conditioned on retrieved anchors, a hypernetwork generates lightweight Low-Rank Adaptation (LoRA) adapters. This ensures parameter efficiency, allowing the model to adapt to each task and object dynamically. Additionally, we formulate an alignment loss that aligns semantic discrepancies in the feature space with functional changes in the parameter space, thereby constraining LoRA adapters to remain focused on the current task and object. Extensive experiments on VQA v2 and NExT-QA under both standard and compositional settings demonstrate the superiority of HyLoVQA over prior state-of-the-art methods.

Visual Question Answering (VQA), as the representative multimodal task, serves as a key benchmark for evaluating the reasoning capabilities of Multimodal Large Language Models (MLLMs). However, existing evaluations largely rely on static datasets and accuracy-based metrics, which fail to capture robustness, consistency, and generalization. Inspired by Metamorphic Testing (MT), we propose Metamorphic Robustness Assessment (MetaRA), a testing framework that employs Metamorphic Relations (MRs) to systematically probe vulnerabilities in MLLM-based VQA systems. MetaRA generates controlled variations of image-question inputs based on specific MRs and evaluates models across diverse conditions. Applying MetaRA to multiple MLLM-based VQA models across different tasks reveals nuanced failure patterns, including sensitivity to linguistic perturbations, over-reliance on superficial visual cues, and deeper weaknesses in multimodal reasoning. Experimental results demonstrate that MetaRA provides richer diagnostic insights than conventional accuracy metrics, exposing failure modes that remain hidden under standard benchmarks. Overall, this work highlights the need for systematic robustness evaluation in VQA and positions metamorphic assessment as a scalable, model-agnostic approach toward trustworthy multimodal AI.

Recent advances in vision-language models (VLMs) have achieved impressive results on standard image-text tasks, yet their potential for visual procedure question answering (VP-QA) remains largely unexplored. VP-QA presents unique challenges where users query next-step actions by uploading images for intermediate states of complex procedures. To systematically evaluate VLMs on this practical task, we propose ProcedureVQA, a novel multimodal benchmark specifically designed for visual procedural reasoning. Through comprehensive analysis, we identify two critical limitations in current VLMs: inadequate cross-modal retrieval of structured procedures given visual states, and misalignment between image sequence granularity and textual step decomposition. To address these issues, we present Chain-of-Procedure (CoP), a hierarchical reasoning framework that first retrieves relevant instructions using visual cues, then performs step refinement through semantic decomposition, and finally generates the next step. Experiments across six VLMs demonstrate CoP's effectiveness, achieving up to 13% absolute improvement over standard baselines.

Long-context modeling is becoming a core capability of modern large vision-language models (LVLMs), enabling sustained context management across long-document understanding, video analysis, and multi-turn tool use in agentic workflows. Yet practical training recipes remain insufficiently explored, particularly for designing and balancing long-context data mixtures. In this work, we present a systematic study of long-context continued pre-training for LVLMs, extending a 7B model from 32K to 128K context with extensive ablations on long-document data. We first show that long-document VQA is substantially more effective than OCR transcription. Building on this observation, our ablations further yield three key findings: i) for sequence-length distribution, balanced data outperforms target-length-focused data (e.g., 128K), suggesting that long-context ability requires generalizable key-information retrieval across various lengths and positions; ii) retrieval remains the primary bottleneck, favoring retrieval-heavy mixtures with modest reasoning data for task diversity; and iii) pure long-document VQA largely preserves short-context capabilities, suggesting that instruction-formatted long data reduces the need for short-data mixing. Based on these findings, we introduce MMProLong, obtained by long-context continued pre-training from Qwen2.5-VL-7B with only a 5B-token budget. MMProLong improves long-document VQA scores by 7.1% and maintains strong performance at 256K and 512K contexts beyond its 128K training window, without additional training. It further generalizes to webpage-based multimodal needle retrieval, long-context vision-text compression, and long-video understanding without task-specific supervision. Overall, our study establishes a practical LongPT recipe and an empirical foundation for advancing long-context vision-language models.

Cancer screening is a reasoning task. A radiologist observes findings, compares them to prior scans, integrates clinical context, and reaches a diagnostic conclusion confirmed by pathology. We present RadThinking, a Visual Question Answering (VQA) dataset that makes this reasoning explicit and trainable. RadThinking releases VQA pairs at three difficulty tiers. Foundation VQAs are atomic perception questions. Single-step reasoning VQAs apply one clinical rule. Compositional VQAs require multi-step chain-of-thought to reach a guideline category such as LI-RADS-5. For every compositional VQA, we release the chain of foundation VQAs that solves it. The chain follows the rules of the governing clinical reporting standard. The dataset spans 20,362 CT scans from 9,131 patients across 43 cancer groups, plus 2,077 verified healthy controls with >1-year follow-up. To our knowledge, RadThinking is the first cancer-screening VQA corpus that stratifies questions by reasoning depth and grounds compositions in clinical reporting standards. The foundation tier supplies atomic perception supervision. The compositional tier supplies chain-of-thought data and verifiable rewards for reinforcement-learning recipes such as DeepSeek-R1 and OpenAI o1. RadThinking enables systematic training and evaluation of whether AI systems can reason about cancer, not merely detect it.

Visual Document Retrieval (VDR) models mostly rely on late interaction architectures, in which documents are represented by a set of local patch embeddings and then matched against query tokens. While efficient, this architecture prioritizes local similarity over global layout structure of documents to estimate relevancy between documents and query. In practice, this leads to errors as relevance originates from layout structure of documents with heterogeneous layouts combining figures, tables, and text. We make document layout learnable without changing inference. We propose a multimodal encoder that augments local patch representations with a global layout embedding, trained via textual descriptions encoding document layout information. Across four ViDoRe-v2 datasets, our model improves over the strongest architecturally comparable ColPali/ColQwen baseline by +2.4 nDCG@5 and +2.3 MAP@5, with statistically significant per-dataset gains over ColQwen.

Mainstream vision-language models (VLMs) fundamentally struggle with severe optical ambiguities, such as reflections and transparent objects, due to the inherent limitations of standard RGB inputs. While polarization imaging captures polarimetric physical parameters that resolve these ambiguities, existing methods are constrained by fixed-format outputs and remain isolated from open-ended reasoning. To bridge this semantic-physical gap, we introduce PolarVLM, the first multimodal framework integrating polarimetric physical parameters into VLMs. By employing a dual-stream architecture and a progressive two-stage training strategy, PolarVLM effectively prevents physical misinterpretations while preserving general visual abilities. Complementing our architecture, we construct PolarVQA, the first benchmark for polarization-aware VQA, featuring 75K physics-grounded instruction-tuning pairs targeting reflective and transparent scenes. Experiments show that PolarVLM surpasses the RGB baseline by 25.4% overall across five evaluation tasks, with remarkable gains of 26.6% in reflection recognition and 34.0% in glass counting, successfully unlocking physics-aware semantic understanding.

Document Visual Question Answering (DocVQA) requires vision-language models to reason not only about what information in a document is relevant to a question, but also where the answer is grounded on the page. Existing DocVQA models entangle question-relevant evidence and answer localization and operate largely as black boxes, offering limited means to verify how predictions depend on visual evidence. We propose CoExVQA, a self-explainable DocVQA framework with a grounded reasoning process through a chain-of-explanation design. CoExVQA first identifies question-relevant evidence, then explicitly localizes the answer region, and finally decodes the answer exclusively from the grounded region. Prediction via CoExVQA's chain-of-explanation enables direct inspection and verification of the reasoning process across modalities. Empirical results show that restricting decoding to grounded evidence achieves SotA explainable DocVQA performance on PFL-DocVQA, improving ANLS by 12% over the current explainable baselines while providing transparent and verifiable predictions.

Video text-based visual question answering (Video TextVQA) aims to answer questions by reasoning over visual textual content appearing in videos. Despite the strong multimodal video understanding capabilities of recent Video-LLMs, their performance on existing Video TextVQA benchmarks remains limited. To better understand this gap, we conduct an upper-bound analysis through frame-wise question answering, counting a sample as correct if any frame yields the right answer, which significantly outperforms direct video-based inference and reveals a substantial performance gap. The results suggest that the primary bottleneck lies in the localization of key question-relevant evidence, rather than in reasoning capacity itself. Building on this insight, we propose a question-guided agent framework that explicitly anchors the relevant keyframes before answering. The approach operates effectively in a training-free setting and consistently surpasses direct video inference. With additional supervised fine-tuning (SFT) and reinforcement learning (RL), it achieves an average improvement of +12.12 in accuracy and +11.15 in ANLS across benchmarks, establishing new state-of-the-art results. Our study underscores the critical role of explicit keyframe anchoring for advancing Video TextVQA. The code will be publicly released.

With advances in multimodal research and deep learning, Multimodal Large Language Models (MLLMs) have emerged as a powerful paradigm for a wide range of multimodal tasks. As a core problem in vision-language research, Visual Question Answering (VQA) has increasingly employed MLLMs to improve performance, particularly in open-domain settings where external knowledge is essential. In this work, we aim to further enhance retrieval-based VQA by more effectively integrating MLLMs with structured reasoning and knowledge acquisition. We introduce a logical prompting strategy that fuses Chain-of-Thought (CoT) reasoning with Visual Question Decomposition (VQD), termed CoVQD, to guide retrieval toward more accurate and relevant knowledge for MLLM inference. Building on this idea, we propose a new framework, CoVQD-guided RAG (CgRAG), which enables MLLMs to access more comprehensive and coherent external knowledge while benefiting from structured visual-text reasoning guidance, thereby improving generalization and reliability in complex cross-domain VQA scenarios. Extensive experiments on E-VQA, InfoSeek, and OKVQA benchmarks demonstrate the effectiveness of the proposed method.

Multidimensional human understanding is essential for real-world applications such as film analysis and virtual digital humans, yet current LVLM benchmarks largely focus on single-task settings and lack fine-grained, human-centric evaluation. In this work, we introduce MHPR, a comprehensive benchmark for joint perception-reasoning over human-centric scenes spanning individual, multi-person, and human-object interaction dimensions. MHPR comprises a multi-level data design-Captioned Raw Data (C-RD), Supervised Fine-Tuning Data (SFT-D), Reinforcement Learning Data (RL-D), and Test Data (T-D)-together with an automated caption/VQA generation pipeline (ACVG) that performs category-wise attribute decomposition, attribute-specific rewriting, and multi-model voting to ensure high-quality, scalable annotations. We evaluate state-of-the-art vision-language models on fine-grained attributes (appearance, clothing, pose, parts) and high-level semantics (social relations, action semantics, spatial relations, intent and functionality). Our findings show that: 1) format-aligned SFT data substantially improves instruction following and stability; 2) challenge-focused RL data derived from bad-case analysis further enhances perception and reasoning on difficult instances; and 3) training Qwen2.5-VL-7B with MHPR yields significant gains, achieving near-parity with considerably larger models. We release ACVG and MHPR to facilitate reproducible, extensible research on human-centric perception and reasoning.

Evaluating generative video models remains an open problem. Reference-based metrics such as Structural Similarity Index Measure (SSIM) and Peak Signal to Noise Ratio (PSNR) reward pixel fidelity over semantic correctness, while Frechet Video Distance (FVD) favors distributional textures over physical plausibility. Binary Visual Question Answering (VQA) based benchmarks like VBench~2.0 are prone to yes-bias and rely on low-resolution auditors that miss temporal failures. Moreover, their prompts target a single dimension at a time, multiplying the number of videos required while still not guaranteeing reliable results. WorldJen addresses these limitations directly. Binary VQA is replaced with Likert-scale questionnaires graded by a VLM that receives frames at native video resolution. Video generation costs are addressed by using adversarially curated prompts that are designed to exercise up to 16 quality dimensions simultaneously. The framework is built around two interlocking contributions. First, A blind human preference study is conducted, accumulating (2,696 pairwise annotations from 7 annotators with 100% pair coverage over 50 of the curated prompts $\times$ 6 state-of-the-art video models. A mean inter-annotator agreement of 66.9% is achieved and the study establishes a human ground-truth Bradley-Terry (BT) rating with a three-tier structure. Second, A VLM-as-a-judge evaluation engine using prompt-specific, dimension-specific Likert questionnaires (10 questions per dimension, 47,160 scored responses) judges the videos and reproduces the human-established three-tier BT rating structure independently. The VLM achieves a Spearman $\hatρ=1.000,~p=0.0014$ that is interpreted as tier agreement with the human results. Six focused ablation studies validate the robustness of the VLM evaluation framework. Project page: https://moonmath.ai/worldjen/

Iterative Retrieval-Augmented Generation (iRAG) has emerged as a powerful paradigm for answering complex multi-hop questions by progressively retrieving and reasoning over external documents. However, current systems predominantly operate on parsed text, which creates two critical bottlenecks: (1) \textit{Coarse-grained attribution}, where users are burdened with manually locating evidence within lengthy documents based on vague text-level citations; and (2) \textit{Visual semantic loss}, where the conversion of visually rich documents (e.g., slides, PDFs with charts) into text discards spatial logic and layout cues essential for reasoning. To bridge this gap, we present \textbf{Chain of Evidence (CoE)}, a retriever-agnostic visual attribution framework that leverages Vision-Language Models to reason directly over screenshots of retrieved document candidates. CoE eliminates format-specific parsing and outputs precise bounding boxes, visualizing the complete reasoning chain within the retrieved candidate set. We evaluate CoE on two distinct benchmarks: \textbf{Wiki-CoE}, a large-scale dataset of structured web pages derived from 2WikiMultiHopQA, and \textbf{SlideVQA}, a challenging dataset of presentation slides featuring complex diagrams and free-form layouts. Experiments demonstrate that fine-tuned Qwen3-VL-8B-Instruct achieves robust performance, significantly outperforming text-based baselines in scenarios requiring visual layout understanding, while establishing a retriever-agnostic solution for pixel-level interpretable iRAG. Our code is available at https://github.com/PeiYangLiu/CoE.git.

We present M$^3$-VQA, a novel knowledge-based Visual Question Answering (VQA) benchmark, to enhance the evaluation of multimodal large language models (MLLMs) in fine-grained multimodal entity understanding and complex multi-hop reasoning. Unlike existing VQA datasets that focus on coarse-grained categories and simple reasoning over single entities, M$^3$-VQA introduces diverse multi-entity questions involving multiple distinct entities from both visual and textual sources. It requires models to perform both sequential and parallel multi-hop reasoning across multiple documents, supported by traceable, detailed evidence and a curated multimodal knowledge base. We evaluate 16 leading MLLMs under three settings: without external knowledge, with gold evidence, and with retrieval-augmented input. The poor results reveal significant challenges for MLLMs in knowledge acquisition and reasoning. Models perform poorly without external information but improve markedly when provided with precise evidence. Furthermore, reasoning-aware agentic retrieval surpasses heuristic methods, highlighting the importance of structured reasoning for complex multimodal understanding. M$^3$-VQA presents a more challenging evaluation for advancing the multimodal reasoning capabilities of MLLMs. Our code and dataset are available at https://github.com/CASIA-IVA-Lab/M3VQA.

Asking inquisitive questions while reading, and looking for their answers, is an important part in human discourse comprehension, curiosity, and creative ideation, and prior work has investigated this in text-only scenarios. However, in scientific or research papers, many of the critical takeaways are conveyed through both figures and the text that analyzes them. While scientific visualizations have been used to evaluate Vision-Language Models (VLMs) capabilities, current benchmarks are limited to questions that focus simply on extracting information from them. Such questions only require lower-level reasoning, do not take into account the context in which a figure appears, and do not reflect the communicative goals the authors wish to achieve. We generate inquisitive questions that reach the depth of questions humans generate when engaging with scientific papers, conditioned on both the figure and the paper's context, and require reasoning across both modalities. To do so, we extend the linguistic theory of Questions Under Discussion (QUD) from being text-only to multimodal, where implicit questions are raised and resolved as discourse progresses. We present MQUD, a dataset of research papers in which such questions are made explicit and annotated by the original authors. We show that fine-tuning a VLM on MQUD shifts the model from generating generic low-level visual questions to content-specific grounding that requires a high-level of multimodal reasoning, yielding higher-quality, more visually grounded multimodal QUD generation.

Recent advances in representation learning have shown that hyperbolic geometry can offer a more expressive alternative to the Euclidean embeddings used in CLIP models, capturing hierarchical structures and leading to better-organized representations. However, current hyperbolic CLIP variants are trained entirely from scratch, which is computationally expensive and resource-intensive. In this work, we propose HAC (Hyperbolic Adaptation of CLIP), a parameter-efficient framework that enables pretrained CLIP models to transition into hyperbolic space via lightweight fine-tuning. We apply HAC to Visual Question Answering (VQA), where models must interpret visual elements and align them with textual queries. Notably, HAC's training is performed on a dataset with no overlap with any VQA benchmark, resulting in a strict zero-shot evaluation paradigm that underscores HAC's task-agnostic adaptability. We evaluate HAC across a diverse suite of VQA benchmarks spanning General, Reasoning, and OCR categories. Both HAC-S (small) and HAC-B (medium) consistently surpass Euclidean baselines and prior hyperbolic approaches, with HAC-B delivering up to a +1.9 point average improvement over CLIP-B on reasoning-intensive tasks. Our code is available at https://github.com/fdibiton/HAC

Stochastic sampling strategies are widely adopted in large language models (LLMs) to balance output coherence and diversity. These heuristics are often inherited in Multimodal LLMs (MLLMs) without task-specific justification. However, we contend that stochastic decoding can be suboptimal for Visual Question Answering (VQA). VQA is a closed-ended task with head-heavy answer distributions where uncertainty is usually epistemic, arising from missing or ambiguous visual evidence rather than plausible continuations. In this work, we provide a theoretical formalization of the relationship between model calibration and predictive accuracy, and derive the sufficient conditions for greedy decoding optimality. Extensive experiments provide empirical evidence for the superiority of greedy decoding over stochastic sampling across multiple benchmarks. Furthermore, we propose Greedy Decoding for Reasoning Models, which outperforms both stochastic sampling and standard greedy decoding in multimodal reasoning scenarios. Overall, our results caution against naively inheriting LLMs decoding heuristics in MLLMs and demonstrate that greedy decoding can be an efficient yet strong default for VQA.

PDF documents contain critical visual elements such as figures, tables, and forms whose accurate extraction is essential for document understanding and multimodal retrieval-augmented generation (RAG). Existing PDF parsers often miss complex visuals, extract non-informative artifacts (e.g., watermarks, logos), produce fragmented elements, and fail to reliably associate captions with their corresponding elements, which degrades downstream retrieval and question answering. We present a lightweight and production level PDF parsing framework that can accurately detect visual elements and associates captions using a combination of spatial heuristics, layout analysis, and semantic similarity. On popular benchmark datasets and internal product data, the proposed solution achieves $\geq96\%$ visual element detection accuracy and $93\%$ caption association accuracy. When used as a preprocessing step for multimodal RAG, it significantly outperforms state-of-the-art parsers and large vision-language models on both internal data and the MMDocRAG benchmark, while reducing latency by over $2\times$. We have deployed the proposed system in challenging production environment.

A common approach to question answering with retrieval-augmented generation (RAG) is to concatenate documents into a single context and pass it to a language model to generate an answer. While simple, this strategy can obscure the contribution of individual documents, making attribution difficult and contributing to the lost-in-the-middle'' effect, where relevant information in long contexts is overlooked. Concatenation also scales poorly: computational cost grows quadratically with context length, a problem that becomes especially severe when the context includes visual data, as in visual question answering. Attempts to mitigate these issues by limiting context length can further restrict performance by preventing models from benefiting from the improved recall offered by deeper retrieval. We propose Bayesian Ensemble Retrieval-Augmented Generation (BERAG), along with Bayesian Ensemble Fine-Tuning (BEFT), as a RAG framework in which language models are conditioned on individual retrieved documents rather than a single combined context. BERAG treats document posterior probabilities as ensemble weights and updates them token by token using Bayes' rule during generation. This approach enables probabilistic re-ranking, parallel memory usage, and clear attribution of document contribution, making it well-suited for large document collections. We evaluate BERAG and BEFT primarily on knowledge-based visual question answering tasks, where models must reason over long, imperfect retrieval lists. The results show substantial improvements over standard RAG, including strong gains on Document Visual Question Answering and multimodal needle-in-a-haystack benchmarks. We also demonstrate that BERAG mitigates the lost-in-the-middle'' effect. The document posterior can be used to detect insufficient grounding and trigger deflection, while document pruning enables faster decoding than standard RAG.

Despite significant progress in Multi-modal Large Language Models (MLLMs), their clinical reasoning capacity for multi-modal diagnosis remains largely unexamined. Current benchmarks, mostly single-modality data, can't evaluate progressive reasoning and cross-modal integration essential for clinical practice. We introduce the Cross-Modality Progressive Clinical Reasoning (X-PCR) benchmark, the first comprehensive evaluation of MLLMs through a complete ophthalmology diagnostic workflow, with two reasoning tasks: 1) a six-stage progressive reasoning chain spanning image quality assessment to clinical decision-making, and 2) a cross-modality reasoning task integrating six imaging modalities. The benchmark comprises 26,415 images and 177,868 expert-verified VQA pairs curated from 51 public datasets, covering 52 ophthalmic diseases. Evaluation of 21 MLLMs reveals critical gaps in progressive reasoning and cross-modal integration. Dataset and code: https://github.com/CVI-SZU/X-PCR.

Change visual question answering (Change VQA) addresses the problem of answering natural-language questions about semantic changes between bi-temporal remote sensing (RS) images. Although vision-language models (VLMs) have recently been studied for temporal RS image understanding, Change VQA remains underexplored in the context of modern multimodal models. In this letter, we revisit the CDVQA benchmark using recent Qwen models under a unified low-rank adaptation (LoRA) setting. We compare Qwen3-VL, which follows a structured vision-language pipeline with multi-depth visual conditioning and a full-attention decoder, with Qwen3.5, a native multimodal model that combines a single-stage alignment with a hybrid decoder backbone. Experimental results on the official CDVQA test splits show that recent VLMs improve over earlier specialized baselines. They further show that performance does not scale monotonically with model size, and that native multimodal models are more effective than structured vision-language pipelines for this task. These findings indicate that tightly integrated multimodal backbones contribute more to performance than scale or explicit multi-depth visual conditioning for language-driven semantic change reasoning in RS imagery.

Despite recent progress in multimodal large language models (MLLMs), reliable visual question answering in aerial scenes remains challenging. In such scenes, task-critical evidence is often carried by small objects, explicit quantities, coarse locations, and inter-object relations, whereas conventional dense visual-token representations are not well aligned with these structured semantics. To address this interface mismatch, we propose AeroRAG, a scene-graph-guided multimodal retrieval-augmented generation framework for visual question answering. The framework first converts an input image into structured visual knowledge, including object categories, quantities, spatial locations, and semantic relations, and then retrieves query-relevant semantic chunks to construct compact prompts for a text-based large language model. Rather than relying on direct reasoning over dense visual tokens, our method introduces a more explicit intermediate interface between perception and language reasoning. Experiments on the AUG aerial dataset and the general-domain VG-150 benchmark show consistent improvements over six strong MLLM baselines, with the largest gains observed in dense aerial scenes and relation-sensitive reasoning. We further evaluate the framework on VQAv2 to verify that the proposed interface remains compatible with standard visual reasoning settings. These results suggest that structured retrieval is a practical design direction for deployment-oriented and grounded visual reasoning systems.

Manual annotation of high-quality visual question answering with grounding (VQA-G) datasets, which pair visual questions with evidential grounding, is crucial for advancing vision-language models (VLMs), but remains unscalable. Existing automated methods are often hindered by two key issues: (1) inconsistent data fidelity due to model hallucinations; (2) brittle verification mechanisms based on simple heuristics. To address these limitations, we introduce AutoVQA-G, a self-improving agentic framework for automated VQA-G annotation. AutoVQA-G employs an iterative refinement loop where a Consistency Evaluation module uses Chain-of-Thought (CoT) reasoning for fine-grained visual verification. Based on this feedback, a memory-augmented Prompt Optimization agent analyzes critiques from failed samples to progressively refine generation prompts. Our experiments show that AutoVQA-G generates VQA-G datasets with superior visual grounding accuracy compared to leading multimodal LLMs, offering a promising approach for creating high-fidelity data to facilitate more robust VLM training and evaluation. Code: https://github.com/rohnson1999/AutoVQA-G

Visual Question Answering (VQA) requires models to identify the correct answer options based on both visual and textual evidence. Recent Mixture-of-Experts (MoE) methods improve option reasoning by grouping similar concepts or routing based on examples. However, unstable routing can lead to inconsistent expert selection in the same question type, while overly stable routing may reduce flexibility. To address this, we propose Concept-Guided Routing framework (CoGR-MoE), which incorporates semantics of the answer options to guide expert selection in the training phase. Next, option features are used to reweight the selected experts, producing discriminative representations for each candidate option. These option-level representations are further used for option comparison and optimized via contrastive learning. The experimental results indicate that CoGR-MoE delivers strong performance across multiple VQA tasks, demonstrating the effectiveness of our approach.

As AI-assisted video creation becomes increasingly practical, instruction-guided video editing has become essential for refining generated or captured footage to meet professional requirements. Yet the field still lacks both a large-scale human-annotated dataset with complete editing examples and a standardized evaluator for comparing editing systems. Existing resources are limited by small scale, missing edited outputs, or the absence of human quality labels, while current evaluation often relies on expensive manual inspection or generic vision-language model judges that are not specialized for editing quality. We introduce VEFX-Dataset, a human-annotated dataset containing 5,049 video editing examples across 9 major editing categories and 32 subcategories, each labeled along three decoupled dimensions: Instruction Following, Rendering Quality, and Edit Exclusivity. Building on VEFX-Dataset, we propose VEFX-Reward, a reward model designed specifically for video editing quality assessment. VEFX-Reward jointly processes the source video, the editing instruction, and the edited video, and predicts per-dimension quality scores via ordinal regression. We further release VEFX-Bench, a benchmark of 300 curated video-prompt pairs for standardized comparison of editing systems. Experiments show that VEFX-Reward aligns more strongly with human judgments than generic VLM judges and prior reward models on both standard IQA/VQA metrics and group-wise preference evaluation. Using VEFX-Reward as an evaluator, we benchmark representative commercial and open-source video editing systems, revealing a persistent gap between visual plausibility, instruction following, and edit locality in current models. Our project page is https://xiangbogaobarry.github.io/VEFX-Bench/.

Adapting pre-trained vision-language models (VLMs) for robotic control requires injecting high-magnitude continuous gradients from a flow-matching action expert into a backbone trained exclusively with cross-entropy. This cross-modal gradient asymmetry - the spectral dimensionality mismatch between low-rank MSE regression gradients and the high-dimensional semantic manifold sculpted by CE pre-training, causes rapid, severe erosion of the VLM's visual-question-answering (VQA) capability. Industry-standard defences either sever the gradient pathway entirely via stop gradient, discarding the rich continuous supervision, or restrict parameter capacity through low-rank adapters (LoRA) that constrain the rank of updates but not their direction, and thus still overwrite the pre-trained manifold. We introduce AEGIS (Anchor-Enforced Gradient Isolation System): a buffer-free, layer-wise orthogonal gradient projection framework that enables direct continuous MSE learning while preserving the pre-trained VQA manifold - without any co-training data or replay buffer. AEGIS pre-computes a static Gaussian reference anchor from masked VQA forward passes across all transformer layers, then at each training step constructs a Wasserstein-2 transport penalty that generates an anchor restoration gradient. A sequential dual-backward decomposes the task and anchor gradients; for each transformer layer, AEGIS applies a single Gram-Schmidt orthogonal projection that bends the task gradient away from the destructive direction while preserving its constructive content. The projection sheds less than 1% of gradient energy on average, yet eliminates the cumulative activation drift that drives severe forgetting.

In continual visual question answering (VQA), existing Continual Learning (CL) methods are mostly built for symmetric, unimodal architectures. However, modern Vision-Language Models (VLMs) violate this assumption, as their trainable components are inherently asymmetric. This structural mismatch renders VLMs highly prone to catastrophic forgetting when learning from continuous data streams. Specifically, the asymmetry causes standard global regularization to favor the massive language decoder during optimization, leaving the smaller but critical visual projection layers highly vulnerable to interference. Consequently, this localized degradation leads to a severe loss of compositional reasoning capabilities. To address this, we propose Asymmetric Information Masking (AIM), which balances stability and plasticity by applying targeted masks based on modality-specific sensitivity. Experiments on VQA v2 and GQA under continual VQA settings show that AIM achieves state-of-the-art performance in both Average Performance (AP) and Average Forgetting (AF), while better preserving generalization to novel skill-concept compositions.

Knowledge-based visual question answering (KB-VQA) requires vision-language models to understand images and use external knowledge, especially for rare entities and long-tail facts. Most existing retrieval-augmented generation (RAG) methods adopt a fixed pipeline that sequentially retrieves information, filters it, and then produces an answer. Such a design makes it difficult to adapt to diverse question types. Moreover, it separates retrieval from reasoning, making it hard for the model to decide when to search, how to refine queries, or when to stop. As a result, the retrieved evidence is often poorly aligned with the question. To address these limitations, we reformulate KB-VQA as a search-agent problem and model the solving process as a multi-step decision-making procedure. At each step, the agent selects one of four actions-Answer, Image Retrieval, Text Retrieval, and Caption-based on its current information state. We further design an automated pipeline to collect multi-step trajectories that record the agent's reasoning process, tool usage, and intermediate decisions. These trajectories are then used as supervision for fine-tuning. Experiments on InfoSeek and E-VQA demonstrate that our method achieves state-of-the-art performance, consistently outperforming prior baselines and confirming the effectiveness of our framework.

Knowledge-Based Visual Question Answering (KB-VQA) requires models to answer questions about an image by integrating external knowledge, posing significant challenges due to noisy retrieval and the structured, encyclopedic nature of the knowledge base. These characteristics create a distributional gap from pretrained multimodal large language models (MLLMs), making effective reasoning and domain adaptation difficult in the post-training stage. In this work, we propose \textit{Wiki-R1}, a data-generation-based curriculum reinforcement learning framework that systematically incentivizes reasoning in MLLMs for KB-VQA. Wiki-R1 constructs a sequence of training distributions aligned with the model's evolving capability, bridging the gap from pretraining to the KB-VQA target distribution. We introduce \textit{controllable curriculum data generation}, which manipulates the retriever to produce samples at desired difficulty levels, and a \textit{curriculum sampling strategy} that selects informative samples likely to yield non-zero advantages during RL updates. Sample difficulty is estimated using observed rewards and propagated to unobserved samples to guide learning. Experiments on two KB-VQA benchmarks, Encyclopedic VQA and InfoSeek, demonstrate that Wiki-R1 achieves new state-of-the-art results, improving accuracy from 35.5% to 37.1% on Encyclopedic VQA and from 40.1% to 44.1% on InfoSeek. The project page is available at https://artanic30.github.io/project_pages/WikiR1/.

Document Visual Question Answering (DocVQA) enables end-to-end reasoning grounded on information present in a document input. While recent models have shown impressive capabilities, they remain vulnerable to adversarial attacks. In this work, we introduce a novel attack scenario that aims to forge document content in a visually imperceptible yet semantically targeted manner, allowing an adversary to induce specific or generally incorrect answers from a DocVQA model. We develop specialized attack algorithms that can produce adversarially forged documents tailored to different attackers' goals, ranging from targeted misinformation to systematic model failure scenarios. We demonstrate the effectiveness of our approach against two end-to-end state-of-the-art models: Pix2Struct, a vision-language transformer that jointly processes image and text through sequence-to-sequence modeling, and Donut, a transformer-based model that directly extracts text and answers questions from document images. Our findings highlight critical vulnerabilities in current DocVQA systems and call for the development of more robust defenses. We release our open source code at https://github.com/pralab/adv-docVQA.

Previous works identify sensitivity to option order as a key issue in multiple-choice VQA (MC-VQA) evaluation and propose protocols to mitigate this effect. We show that such mitigation is insufficient to ensure the validity of MC-VQA as a reliable benchmark for Multimodal Large Language Model (MLLMs): performance remains highly sensitive to semantically neutral prompt format choices that are not controlled by current benchmarks. In a large-scale study spanning seven MLLMs and five MC-VQAs datasets, we find frequent rank reversals even under order-invariant evaluation. These reversals arise when we systematically vary option ID sets, delimiters, and separators, yielding 48 semantically equivalent prompt formats. Mechanistic analyses trace this instability to low-level language modeling effects: tokenizer-induced fusion or removal of option ID tokens introduces corrupted option ID tokens into the input sequence, while the choice of option ID sets directly affects the reliability of attention patterns for option selection. Accordingly, MC-VQA rankings correlate weakly with open-ended evaluation, indicating that MC-VQA reflects option-selection dynamics in addition to multimodal reasoning. These findings identify prompt formatting as a major, previously under-controlled confounder in MC-VQA benchmarking and motivate evaluation protocols that explicitly control prompt format sensitivity.

Knowledge-based Visual Question Answering (KBVQA) necessitates external knowledge incorporation beyond cross-modal understanding. Existing KBVQA methods either utilize implicit knowledge in multimodal large language models via in-context learning or explicit knowledge via retrieval augmented generation. However, their reasoning processes remain implicit, without explicit multi-step trajectories. To address this gap, we propose a Self-Encouraged Hindsight Distillation Reasoning (HinD) framework, aiming at eliciting reasoning ability inside the MLLM by constructing a Hindsight Teacher with privileged information to teach the Foresight Student. First, we construct the Hindsight Teacher by prompting the MLLM with the reasoning target as privileged information to complete the reasoning process, obtaining Hindsight-Zero training data. Then, the Foresight Student, without knowing the answer, learns the golden trajectories from Hindsight in two ways: (1) Hindsight Distillation Fine-Tuning to self-distill the Hindsight-Zero into a modularized Chain-of-Thought Generator and a Knowledge Generator for sequential steps and discrete facts generation, respectively; (2) Knowledge Encouragement Preference Optimization to encourage the under-confident but relevant knowledge inside the MLLM and suppress the over-confident but irrelevant one. Experiments on OK-VQA and A-OKVQA validate the effectiveness of HinD, showing that HinD with 7-8B MLLM achieves superior performance without commercial model APIs or retrieved knowledge.

Large Language Models (LLMs) with multimodal capabilities have revolutionized vision-language tasks, but their deployment often requires huge memory and computational resources. Post-training quantization (PTQ) has successfully compressed language models to as low as 1-bit precision, its effectiveness for multimodal LLMs (MLLMs) remains unexplored. In this paper, we present the first method for ultra-low-bit (<4-bit) quantization of MLLMs. Our analysis reveals that multimodal tokens and intermediate layer activations produced by them exhibit significantly higher entropy compared to text tokens, indicating greater functional complexity that makes MLLMs less tolerant to ultra-low bit quantization. However, this entropy varies significantly across layers, with some layers producing lower-entropy activation distributions that we empirically show can better tolerate ultra-low bit quantization. Existing PTQ methods optimize weight quantization within each layer but apply the same target precision uniformly, ignoring this variation in complexity across layers. Building on this insight, we propose LUQ: Layerwise Ultra-Low Bit Quantization, which characterizes each transformer layer's functional complexity via its output activation entropy and selectively applies ultra-low bit quantization to layers encoding simpler, more compressible functions. We also show that multimodal calibration (image and text tokens) boosts VQA performance in the ultra-low bit regime. Evaluated on LLaVA-1.5 and Qwen-2.5-VL across 9 VQA benchmarks, LUQ models use 40% and 31% less memory than their 4-bit counterparts while exhibiting less than 10% degradation on MME.