Panel captioning on BioSci-Fig-Cap
5.53BLEU-4FigEx2-8B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| FigEx2-8BProtocol=Fine-tuned2026.01 | 5.53 | 26.35 | 18.78 | 87.24 | |
| Qwen3-VL-8BProtocol=Fine-tuned2026.01 | 5.21 | 25.69 | 18.27 | 87 | |
| gpt-5-miniProtocol=Pre-trained2026.01 | 4.44 | 23.03 | 16.2 | 86.67 | |
| gpt-5.2Protocol=Pre-trained2026.01 | 4.07 | 23.85 | 14.63 | 87.12 | |
| Llama3.2-11BProtocol=Fine-tuned2026.01 | 3.88 | 21.44 | 15.79 | 86.33 | |
| LLaVA-NeXTProtocol=Fine-tuned2026.01 | 3.36 | 20.2 | 12.74 | 86.22 | |
| Llama3.2-90BProtocol=Pre-trained2026.01 | 2.94 | 18.35 | 11.8 | 85.96 | |
| LLaVA-MedProtocol=Fine-tuned2026.01 | 2.91 | 17.57 | 11.32 | 85.78 | |
| gpt-5-nanoProtocol=Pre-trained2026.01 | 2.84 | 16.97 | 9.76 | 85.79 | |
| Qwen2.5-VL-72BProtocol=Pre-trained2026.01 | 1.71 | 23.62 | 14.74 | 82.23 |