MEDVQA-GI-2026-Task1-VQA

Models fine-tuned for ImageCLEFmed MEDVQA-GI 2026 — Subtask 1: Clinically Relevant VQA.

Results

Model	BLEU-4	ROUGE-L	METEOR	Token-F1	BERTScore-F1
Qwen2.5-VL-Base	0.0719	0.3388	0.2655	0.2563	0.3631
Qwen2.5-VL-KvasirFT	0.1326	0.4499	0.3760	0.4170	0.5107
Qwen2.5-VL-TransfFT	0.1413	0.4532	0.3837	0.4249	0.5146
MedGemma-KvasirFT	0.1520	0.4594	0.4057	0.4402	0.5133
Hybrid	0.1389	0.4401	0.3764	0.4074	0.4955

Best model: MedGemma-KvasirFT