UNCC at MedGenVidQA 2026: Structured Temporal Grounding for Medical Video Question Answering

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Health & Medical Research, Life Sciences & Biology · Depth: Expert, quick

Summary

The University of North Carolina at Charlotte (UNCC) participated in the MedGenVidQA 2026 Task C, focusing on visual answer localization within medical videos. Their framework processes a video and a question to identify the precise start and end times of the visual answer. It utilizes timestamped automatic speech recognition (ASR) as a proposal source, generating detailed transcript tables, phase maps, and various candidate windows. The system employs schema-constrained ranking and selective key-frame checks, culminating in a deterministic validation pass for the final JSON output. This approach ranks among bounded candidate intervals rather than generating arbitrary timestamps. UNCC's best submission secured fifth place out of six participant systems, achieving an IoU@0.3 of 62.50, IoU@0.5 of 36.25, IoU@0.7 of 22.50, and a mean IoU (mIoU) of 42.57. The performance pattern indicates that coarse temporal retrieval proved more robust than exact start-end localization.

Key takeaway

For NLP Engineers and Research Scientists developing medical video question answering systems, you should explore structured temporal grounding approaches. The UNCC framework demonstrates that using timestamped ASR as a proposal source, combined with schema-constrained ranking over bounded candidate intervals, can achieve competitive results in visual answer localization. Prioritize robust coarse temporal retrieval over strict start-end localization, as this approach proved more reliable. This method offers a traceable output, enhancing system interpretability and debugging.

Key insights

The UNCC framework for medical video QA uses structured temporal grounding with ASR proposals for visual answer localization.

Principles

• ASR can serve as a proposal source, not just a final label.
• Coarse temporal retrieval may outperform strict localization.
• Structured ranking over bounded intervals improves precision.

Method

The framework generates transcript tables, phase maps, lexical/dense candidate windows, and uses schema-constrained ranking with selective key-frame checks and a deterministic validation pass.

In practice

• Implement ASR for initial temporal segment proposals.
• Prioritize coarse temporal grounding over exact boundaries.
• Use schema-constrained ranking for candidate intervals.

Topics

• Medical Video QA
• Temporal Grounding
• ASR Proposals
• Visual Answer Localization
• Schema-Constrained Ranking
• BioNLP

Best for: AI Scientist, Research Scientist, NLP Engineer

Related on AIssential

• Natural-Language Temporal Grounding in Hour-Long Videos is a Search Problem: A Benchmark and Empirical Decomposition — For hour-long videos, natural-language temporal grounding is primarily a search problem, not a recognition bottleneck for Video-LLMs.
• EG-VQA: Benchmarking Verifiable Video Question Answering with Grounded Temporal Evidence — Video-LLMs need explicit evidence grounding and structured supervision, not just scaling, for robust and interpretable video question answering.
• MedSIGHT: Towards Grounded Visual Comprehension in Medical Large Vision-Language Models — MedSIGHT unifies medical visual comprehension and pixel-level segmentation via region-centric tokens and a modality-aware codebook.
• Temporal-Aware Reasoning Optimization for Video Temporal Grounding — TaRO improves MLLM video temporal grounding by optimizing time-aware reasoning through guided exploration and quality-focused rewards.
• Hierarchical Semantic-Constrained Heterogeneous Graph for Audio-Visual Event Localization — HSCHG improves open-vocabulary audio-visual event localization by combining Euclidean graph modeling with hyperbolic space for hierarchical semantic alignment.
• Grounded SAM 2: From Open-Set Detection to Segmentation and Tracking — Grounded SAM 2 combines language-driven detection with pixel-level segmentation and video tracking for comprehensive visual understanding.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.