4arXiv cs.CL (Computation and Language)·
Retrospective analysis of multimodal VQA systems for clinical GI endoscopy reveals evaluation gaps
A new arXiv paper analyzes nine documented systems from the MediaEval Medico 2025 challenge for GI endoscopy visual question answering, examining design choices around parameter-efficient adaptation, structured reasoning, and explanation quality. The authors find that leaderboard performance gains do not reliably translate to faithful clinical reasoning, while methods enforcing structured reasoning and explicit grounding show more consistent behavior. The paper argues for evaluation frameworks that go beyond lexical overlap, incorporating evidence-linked explanations, leakage-aware data governance, and calibration checks for trustworthy healthcare AI.
Related guides (2)
Related events (8)
Training framework reduces calibration error 60%+ in Medical VQA multimodal LLMs
A new arXiv preprint proposes a finetuning framework to improve verbalized uncertainty calibration in multimodal LLMs applied to Medical Visual Question Answering. The composite loss function combines Brier-style calibration, anchor regularization, contrastive image-text alignment, and KL-based stabilization, evaluated on MedGemma 4B IT and Qwen2-VL 7B Instruct across three medical VQA benchmarks. The method reduces calibration error by 60% or more and improves discrimination by 26% or more while preserving predictive accuracy, outperforming prompting-, sampling-, and training-based baselines.
SHOVIR benchmark exposes vision shortcut learning failures in radiology report generation VLMs
Researchers introduce SHOVIR, a benchmark for detecting 'vision shortcut' behavior in Vision-Language Models applied to Radiology Report Generation (RRG), where models achieve high scores by exploiting learned priors rather than actual image evidence. The benchmark extends MIMIC-CXR and PadChest-GR with per-box CheXpert labels and uses localized occlusion experiments to isolate two failure modes: direct shortcuts (findings persist after visual evidence is removed) and contextual shortcuts (detection degrades when co-occurring pathologies are occluded). Evaluating eight state-of-the-art VLMs, the authors find that high report quality does not correlate with strong spatial grounding, revealing a systematic blind spot in current RRG evaluation protocols.
EG-VQA benchmark exposes gap between answer correctness and evidence grounding in Video-LLMs
Researchers introduce EG-VQA, a benchmark of 2,067 videos and 11,838 QA pairs designed to evaluate Video Large Language Models on both answer correctness and temporal evidence grounding simultaneously. A new metric, Evidence-Grounded F1 (EG-F1), jointly measures temporal alignment and semantic consistency against annotated ground-truth evidence. Experiments show that even strong proprietary models fail to reliably localize supporting evidence, revealing a fundamental gap between surface-level accuracy and faithful reasoning. The authors also propose EG-Reasoner, an evidence-supervised open-source model that achieves competitive results against proprietary systems.
ParaEval framework reduces MCQA benchmark sensitivity to answer phrasing
A new arXiv preprint identifies a systematic flaw in multiple-choice QA benchmarks: log-likelihood scoring conflates surface-form familiarity with actual capability, producing false performance gaps exceeding 2 points between models trained on identical knowledge. The authors propose ParaEval, which queries models with multiple paraphrases per answer option and scores on the most favorable phrasing, reducing the false gap to below 1 point. The effect is confirmed on frontier 70B and 120B open-source models, suggesting widespread benchmark inflation in standard MCQA evaluations.
OpenMedReason: Large-scale multimodal medical reasoning corpus with 450K instances for clinical VLM training
Researchers introduce OpenMedReason, a 450K-instance open multimodal medical reasoning corpus with reasoning traces derived from human-authored biomedical literature rather than synthetic chains of thought. The dataset covers diverse medical imaging modalities and is paired with OpenMedReason-Bench, a held-out benchmark evaluating LVLMs on perception, medical knowledge, and rationale axes. Training with OpenMedReason yields a 20% average VQA accuracy improvement over base models and achieves performance within 4.2% of leading comparable-scale medical VLMs. Both the dataset and code are publicly released.
Test-time scaling for small VLMs on multilingual visual MCQ: conditions matter more than methods
A new arXiv paper examines whether test-time scaling (TTS) transfers to small open vision-language models using EXAMS-V, a multilingual visual multiple-choice benchmark. The study compares self-consistency, describe-then-reason with PRM-guided beam search, and post-hoc selectors across Qwen2.5-VL-7B-Instruct and Qwen3.5-4B. Key findings: prompt parseability and decoding budget (token limit) dominate gains, while elaborate search/verification methods like PRM-guided beam search underperform plain majority vote at 8x the cost. The best configuration achieves 84.1% on ImageCLEF 2026 test split, ranking first on the Visual MCQ leaderboard.
LAVE: Zero-shot VQA Evaluation on Docmatix with LLMs - Do We Still Need Fine-Tuning?
This Hugging Face blog post introduces LAVE (LLM-Assisted Visual Evaluation), a zero-shot VQA evaluation methodology applied to the Docmatix dataset. The post investigates whether large vision-language models can perform document visual question answering without task-specific fine-tuning by leveraging LLM-based evaluation metrics. The analysis probes the gap between zero-shot and fine-tuned performance on document understanding tasks, raising questions about the continued necessity of supervised adaptation for VQA.
ThReadMed-QA benchmark reveals frontier LLMs degrade sharply on medical misconception correction across multi-turn conversations
Researchers introduce ThReadMed-QA, a multi-turn medical dialogue dataset of 2,437 conversation threads (8,204 QA pairs) derived from real AskDocs patient interactions, designed to evaluate LLM ability to detect and correct embedded misconceptions over multiple turns. Evaluation of five LLMs using an LLM-as-a-Judge rubric finds that even frontier models like GPT-5 and Claude Haiku drop from ~85% misconception correction on initial questions to ~50% within two follow-up turns. Oracle analysis attributes much of this degradation to error propagation from prior model outputs rather than context length alone. The findings highlight a significant safety gap in current evaluation frameworks, which do not capture multi-turn conversational dynamics in medical settings.