Elamur Research⌕
☰
home / papers / 2026 / LAMAR-2 at MedGenVidQA 2026: Visual Answer Localization in Medical Videos via Multimodal LLM and Context-Augmented Prompting

LAMAR-2 at MedGenVidQA 2026: Visual Answer Localization in Medical Videos via Multimodal LLM and Context-Augmented Prompting

Paper · 2026 · Watcharitpol Sermsrisuwan, Nopporn Lekuthai, Seksan Yoadsanit et al.

Summary

The paper tackles visual answer localization for the MedGenVidQA 2026 Task C shared task: given a clinical question and an untrimmed medical instructional video, the system must return the start and end times of the segment that answers it. The authors treat this as a generative, multimodal problem. Videos are split at camera cuts with PySceneDetect, and Qwen3-VL describes each segment. These descriptions are aligned with timestamped ASR transcripts to form fused context blocks, which are passed to Gemini-3-Flash together with the raw video.

They compare unimodal baselines, RAG over transcripts with rank-based temporal merging, and several prompting strategies. A "Heuristic Context" prompt works best. It tells the model to use speech only as a rough hint and to anchor boundaries on visible physical movements. This setup ranked first on the leaderboard with an mIoU of 79.55.

The work shows that spoken narration often runs ahead of or lags behind the actual procedure. Countering this text bias is key for precise temporal grounding in clinical video.

Key points

  • Data:
    • Development used 49 MedVidQA videos with 148 QA pairs, a subset of the original 3,010 pairs from 900 videos, because many videos could not be downloaded.
    • The official test set contains 80 questions over 65 videos.
  • Official test scores: mIoU 79.55, IoU@0.3 93.75, IoU@0.5 90.00 and IoU@0.7 77.50 (state of the art on the leaderboard).
  • ASR comparison: Among Parakeet-tdt-0.6b-v2, Whisper large-v3 and Qwen3-ASR-1.7B, Parakeet gave the best transcript-only mIoU (66.03).
  • What failed: Feeding raw transcripts or RAG-retrieved chunks, even with Temporal Reciprocal Rank Fusion, hurt precision. The model tended to align boundaries with spoken keywords rather than the procedure itself.
  • Limitations and release:
    • The method depends on the proprietary Gemini-3-Flash.
    • Evaluation covers English data only.
    • The system is sensitive to poor video quality.
    • Code is released at github.com/biodatlab/medgenvidqa-lamar.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Mahidol UniversityBangkok · Thailand3 works · 2 citations

Also on the byline: Ramathibodi Hospital.

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 1Research subjects and languages the work is about.Vision & language · 657 works

Cite

Cite as Watcharitpol Sermsrisuwan, Nopporn Lekuthai, Seksan Yoadsanit and Titipat Achakulvisut. 2026. LAMAR-2 at MedGenVidQA 2026: Visual Answer Localization in Medical Videos via Multimodal LLM and Context-Augmented Prompting. Proceedings of the BioNLP 2026 (Shared Tasks).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.bionlp-2.31

@inproceedings{sermsrisuwan2026lamar,
  title = {{LAMAR-2 at MedGenVidQA 2026: Visual Answer Localization in Medical Videos via Multimodal LLM and Context-Augmented Prompting}},
  author = {Watcharitpol Sermsrisuwan, Nopporn Lekuthai, Seksan Yoadsanit and Titipat Achakulvisut},
  booktitle = {Proceedings of the BioNLP 2026 (Shared Tasks)},
  year = {2026},
  doi = {10.18653/v1/2026.bionlp-2.31},
  url = {https://doi.org/10.18653/v1/2026.bionlp-2.31}
}

References

24 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Rawan AlSaad, Alaa A. Abd-alrazaq, Sabri Boughorbel, Arfan Ahmed, Max-Antoine Renault, R. Damseh, and Javaid Sheikh. 2024. Multimodal large language models in health care: Applications, challenges, and future outlook. Journal of Medical Internet Research, 26. find
  2. Shuai Bai, Yuxuan Cai, Ruizhe Chen, Keqin Chen, Xionghui Chen, Zesen Cheng, Lianghao Deng, Wei Ding, Chang Gao, Chunjiang Ge, Wenbin Ge, Zhifang Guo, Qidong Huang, Jie Huang, Fei Huang, Binyuan Hui, Shutong Jiang, Zhaohai Li, Mingsheng Li, and 45 others. 2025. Qwen3-vl technical report. arXiv prepri…
  3. Meng Cao, Long Chen, Mike Zheng Shou, Can Zhang, and Yuexian Zou. 2021. On pursuit of designing multi-modal transformer for video grounding. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 9810– 9823, Online and Punta Cana, Dominican Republic. Associa… find
  4. Deepak Gupta, Kush Attal, and Dina Demner-Fushman. 2023. A dataset for medical instructional video classification and question answering. Scientific Data, 10(1):158. find
  5. Deepak Gupta, Kush Attal, and Dina Demner-Fushman. 2024. Towards answering health-related questions from medical videos: Datasets and approaches. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 16399–… find
  6. Deepak Gupta, Collin Scott Campbell, Pedram Golnari, and Dina Demner-Fushman. 2026. Overview of the medgenvidqa 2026 shared task on medical generative video question answering. In Proceedings of the 25th Workshop on Biomedical Language Processing (BioNLP 2026), San Diego, USA. Association for Comput… find
  7. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. In Proceedings of the 34th Internation…
  8. Chaoyu Li, Eun Woo Im, and Pooyan Fazli. 2025. Vidhalluc: Evaluating temporal hallucinations in multimodal large language models for video understanding. In 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 13723– 13733. find
  9. Jiajie Li, Garrett C. Skinner, Gene Yang, Brian R Quaranto, Steven D. Schwaitzberg, Peter C W Kim, and Jinjun Xiong. 2024a. Llava-surg: Towards multimodal surgical assistant via structured surgical video learning. ArXiv, abs/2408.07981. — [arXiv:2408.07981](https://arxiv.org/abs/2408.07981) find
  10. Shutao Li, Bin Li, Bin Sun, and Yixuan Weng. 2024b. Towards visual-prompt temporal answer grounding in instructional video. IEEE Transactions on Pattern Analysis and Machine Intelligence, 46(12):8836– 8853. find
  11. Brandon Castellano and contributors. 2025. PySceneDetect: A video scene cut detection and analysis tool. https://github.com/Breakthrough/ PySceneDetect. Accessed: 2026-04-29. ↗
  12. Kevin Qinghong Lin, Pengchuan Zhang, Joya Chen, Shraman Pramanick, Difei Gao, Alex Jinpeng Wang, Rui Yan, and Mike Zheng Shou. 2023. Univtg: Towards unified video-language temporal grounding. In 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 2782–2792. find
  13. Zhen Chen, Xingjian Luo, Kun Yuan, Jinlin Wu, Danny T. M. Chan, Nassir Navab, Hongbin Liu, Zhen Lei, and Jiebo Luo. 2025. Surgllm: A versatile large multimodal model with spatial focus and temporal NVIDIA. 2024. Parakeet-TDT-0.6B-v2: State-of-the-art speech recognition model. https://huggingface.co… ↗
  14. Zackary Rackauckas. 2024. Rag-fusion: a new take on retrieval-augmented generation. ArXiv, abs/2402.03367. — [arXiv:2402.03367](https://arxiv.org/abs/2402.03367) find
  15. Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. 2022. Robust speech recognition via large-scale weak supervision. arXiv preprint.
  16. Henrique* Schechter Vera, Sahil* Dua, Biao Zhang, Daniel Salz, Ryan Mullins, Sindhu Raghuram Panyam, Sara Smoot, Iftekhar Naim, Joe Zou, Feiyang Chen, Daniel Cer, Alice Lisak, Min Choi, Lucas Gonzalez, Omar Sanseviero, Glenn Cameron, Ian Ballantyne, Kat Black, Kaifeng Chen, and 69 others. 2025. Embe… find
  17. Xian Shi, Xiong Wang, Zhifang Guo, Yongqi Wang, Pei Zhang, Xinyu Zhang, Zishan Guo, Hongkun Hao, Yu Xi, Baosong Yang, Jin Xu, Jingren Zhou, and Junyang Lin. 2026. Qwen3-asr technical report. arXiv preprint arXiv:2601.21337.
  18. K. Nyangoh Timoh, Arnaud Huaulmé, K. Cleary, Myra A. Zaheer, V. Lavoué, D. Donoho, and P. Jan-nin. 2023. A systematic review of annotation for surgical process model analysis in minimally invasive surgery based on video. Surgical Endoscopy, 37:4298 – 4314. find
  19. Guan-Feng Wang, Wenjin Mo, Junyi Wang, Long Bai, Kun Yuan, Ming Hu, Jinlin Wu, Junjun He, Yiming Huang, N. Padoy, Zhen Lei, Hongbin Liu, Nassir Navab, and Hongliang Ren. 2025. Surgvidlm: Towards multi-grained surgical video understanding with large language model. ArXiv, abs/2506.17873. — [arXiv:2506.17873](https://arxiv.org/abs/2506.17873) find
  20. Thomas Winterbottom, Sarah Xiao, Alistair McLean, and N. A. Moubayed. 2020. On modality bias in the tvqa dataset. ArXiv, abs/2012.10210. — [arXiv:2012.10210](https://arxiv.org/abs/2012.10210) find
  21. Jie Wu, Guanbin Li, Xiaoguang Han, and Liang Lin. 2020. Reinforcement learning for weakly supervised temporal grounding of natural language in untrimmed videos. In Proceedings of the 28th ACM International Conference on Multimedia, MM ’20, page 1283–1291, New York, NY, USA. Association for Computing… find
  22. Junbin Xiao, Angela Yao, Yicong Li, and Tat-Seng Chua. 2023. Can i trust your answer? visually grounded video question answering. 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 13204–13214. find
  23. Hao Zhang, Aixin Sun, Wei Jing, and Joey Tianyi Zhou. 2023b. Temporal sentence grounding in videos: A survey and future directions. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(8):10443–10465. find
  24. Xuemei Zhang, Peng Zhao, Jinsheng Ji, Xiankai Lu, and Yilong Yin. 2024. Video corpus moment retrieval via deformable multigranularity feature fusion and adversarial training. IEEE Transactions on Circuits and Systems for Video Technology, 34(8):6686–6698. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sermsrisuwan-2026-lamar-2-medgenvidqa-2026-visual/