Tag
The paper introduces MInfoVisQA, a parallel multilingual benchmark with 8,960 QA configurations across 24 languages, evaluating 25 MLLMs on information-rich visual question answering and revealing substantial cross-language robustness gaps despite similar aggregate accuracy.
AmbiSteer 提出一个轻量框架,通过线性分类器预测输入歧义并条件化激活方向,选择性引导视觉语言模型产生澄清或枚举等歧义感知响应,将适当处理输入的比例从 59.0% 提升至 77.0%。研究发现单独使用引导效果有限,而提示与引导结合效果更佳。
The paper introduces MedJev, a System One vision-language decision model that maps medical images, questions, and candidate answers directly to decisions in a single forward pass without autoregressive generation. Built on Qwen3.5-4B, MedJev-4B achieves a 7.93% absolute accuracy gain and 6.04x end-to-end speedup, reaching state-of-the-art performance among 4B–9B general and medical VLMs.
该论文系统比较了LLM生成的知识引导在知识型VQA中的两种集成策略,发现浅层的决策级晚期融合(39.4% EM)在数据稀缺场景下优于深层的注意力注入(31.8%),并提出了一种基于LLM的数据增强管线,将训练数据扩充十倍以上。
This paper evaluates how Vision Language Models such as ChatGPT, Claude, Gemini, and LLaVA handle VQA questions that violate Grice's maxims through non-essential, ambiguous, or false modifiers, showing diminished performance and differences in pragmatic reasoning compared to humans.
The paper introduces VisKG, a reinforcement learning framework that trains vision-language models to translate images into question-specific knowledge graph representations, improving VQA reasoning efficiency while reducing token usage. A proposed GDPO objective outperforms GRPO-trained counterparts by an average of 2% accuracy across science, math, and general visual reasoning benchmarks.
EviRover reframes visual perception as an agentic process that gathers evidence beyond a single glance, training a 4B perception agent with SFT and agentic RL on newly generated datasets (EviRover-SFT-5K, EviRover-RL-12K) and a human-verified benchmark (EviLens), outperforming its backbone by 30 points and rivaling proprietary models.
The ICDAR2026 competition on multimodal reasoning over documents presented results from 8 teams, highlighting advanced systems that use structured evidence extraction and multi-component orchestration for VQA tasks across diverse document domains.
This paper introduces TestHallVQA, a multi-image VQA benchmark for evaluating Large Vision-Language Models' document-level reasoning under redundant contexts, and proposes a new metric F1-R2 to quantify computational reasoning and robustness.
ReVA introduces a region-aware visual assistant that enhances visually grounded question answering by integrating whole-image and region-level representations, reducing hallucinations in multimodal large language models.
KBMR is an MLLM-based retriever for knowledge-based visual question answering that uses semantic alignment to improve retrieval and VQA accuracy, showing significant gains over CLIP baselines.
This paper introduces Bayesian Data Reweighting, a probabilistic framework that adaptively weights query-document pairs to mitigate false negatives in contrastive training, improving multimodal retrieval across multiple knowledge-based VQA benchmarks.
This paper introduces LedgerMind, a provenance-constrained multimodal agentic reasoning framework that uses a Structured Evidence Ledger to ensure grounded, faithful reasoning in visual question answering, addressing failure patterns like hallucination and over-reasoning.
A new vision-encoder-decoder model is introduced that can process both images and text to generate human-like responses, suitable for tasks like image captioning and visual question answering.
This paper adapts a mixture-of-experts diffusion language model, DiffusionGemma-26B, for interactive radiology report drafting, showing it matches or exceeds autoregressive models in medical VQA with 3.5-4.4x faster decoding and bidirectional infill capabilities.
Proposes ProMSA, a progressive multimodal search agent for knowledge-based visual question answering that adaptively selects search strategies and optimizes through sequence-level reinforcement learning, achieving consistent gains on E-VQA and InfoSeek.
DR-MV3D presents a map-grounded learning framework with dense rewards to improve multi-view 3D visual question answering through global map construction, view-trajectory planning, and egocentric grounding.
Researchers introduce camroll, a dataset of 50 users, 31K images, and 2.5K QA pairs for personal camera roll visual question answering, along with camroll-agent, a conversational AI agent with hierarchical memory for navigating large personalized visual memory. The work highlights gaps in long-context reasoning for personalized visual content compared to standard textual memory approaches.
SuperMemory-VQA is a new egocentric VQA benchmark featuring 52.9 hours of AI-glasses footage and 4,853 QA pairs designed to evaluate AI assistants on long-horizon memory tasks spanning object recall, intent, timelines, and conversations. Benchmarking reveals existing agentic frameworks and LLMs remain far from reliable on these real-world memory challenges.
This paper introduces MechVQA, a dataset with 3.3k high-density mechanical engineering drawings and 21k question-answer pairs, along with the MechVL model that outperforms existing baselines by 7.57 percentage points on the MechVQA total score, advancing multimodal LLM understanding of mechanical drawings.