Structured report generation on PFED5+
67.39BERT ScoreTraMP-LLaMA
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| TraMP-LLaMAVision Encoder=SigLIP-NaViT + SkateFormer, LLM=Qwen2.5-7B2026.06 | 67.39 | 41.88 | 28 | 28.47 | |
| VideoLLaMA3Vision Encoder=SigLIP-NaViT, LLM=Qwen2.5-7B2026.06 | 66.89 | 40.93 | 27.25 | 26.32 | |
| Chat-UniViVision Encoder=CLIP ViT-L/14, LLM=Vicuna-7B2026.06 | 49.13 | 21.33 | 3.26 | 2.6 | |
| LongVUVision Encoder=SigLIP-so400m, LLM=Qwen2-7B2026.06 | 47.65 | 20.64 | 2.26 | 2.2 | |
| LLaVA-VideoVision Encoder=SigLIP-so400m, LLM=Qwen2-7B2026.06 | 35.16 | 18.85 | 1.98 | 1.6 |