Surgical Video Question Answering on EndoVis18-VQA Out-of-template
28.15BLEU-4RL framework over Digital Twin Representations
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| RL framework over Digital Twin Representations2026.06 | 28.15 | 55.9 | 68.45 | 53.4 | |
| SurgViVQALLM=GPT-2, Modality=Video, Training Paradigm=Ours2025.11 | 24.84 | 50.86 | 50.13 | 9.23 | |
| SurgViVQA2026.06 | 24.84 | 50.86 | 50.13 | 9.23 | |
| PitVQALLM=GPT-2, Modality=Image, Training Paradigm=Base2025.11 | 17.57 | 46.87 | 45.49 | 9.73 | |
| PitVQA2026.06 | 17.57 | 46.87 | 45.49 | 9.73 | |
| Qwen3-VLParameters=8B2026.06 | 15.8 | 55.2 | 66.9 | 50.3 | |
| Surgical-LVLM2026.06 | 15 | 54.8 | 66.2 | 49.8 | |
| Qwen3-VLParameters=4B2026.06 | 13.2 | 53.5 | 64.8 | 48.1 | |
| Qwen2.5LLM=-, Modality=Image, Training Paradigm=Zero-Shot2025.11 | 11.92 | 52.77 | 63.72 | 46.9 | |
| Qwen2.5-VLParameters=3B2026.06 | 11.92 | 52.77 | 63.72 | 46.9 | |
| MedGemmaLLM=-, Modality=Image, Training Paradigm=Zero-Shot2025.11 | 7.74 | 40.74 | 61.59 | 37.86 | |
| MedGemmaParameters=4B2026.06 | 7.74 | 40.74 | 61.59 | 37.86 | |
| InternVL3 + LoRALLM=-, Modality=Video, Training Paradigm=FT2025.11 | 6.27 | 28.04 | 37.08 | 56.42 | |
| SurgicalGPTLLM=GPT-2, Modality=Image, Training Paradigm=Base2025.11 | 2.52 | 43.97 | 44.99 | 11.85 | |
| SurgicalGPT2026.06 | 2.52 | 43.97 | 44.99 | 11.85 | |
| SurgViVQALLM=Qwen, Modality=Video, Training Paradigm=Ours2025.11 | 0.6 | 12.33 | 8.1 | 0.99 | |
| InternVL3LLM=-, Modality=Video, Training Paradigm=Zero-Shot2025.11 | 0.57 | 38.73 | 38.73 | 21.99 | |
| VideoLLaMA3LLM=-, Modality=Video, Training Paradigm=Zero-Shot2025.11 | 0.57 | 38.73 | 39.51 | 35.89 | |
| InternVL3Parameters=1B2026.06 | 0.57 | 38.73 | 38.73 | 21.99 | |
| VideoLLaMA3Parameters=2B2026.06 | 0.57 | 38.73 | 39.51 | 35.89 |