Panel Captioning on PhysSci-Fig-Cap (test)
2.89BLEU-4FigEx2-8B
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| FigEx2-8BEvaluation Protocol=Fine-tuned2026.01 | 2.89 | 18.24 | 10.52 | 85.69 | |
| gpt-5-miniEvaluation Protocol=Pre-trained2026.01 | 2.81 | 16.11 | 10.56 | 84.81 | |
| gpt-5.2Evaluation Protocol=Pre-trained2026.01 | 2.77 | 16.71 | 9.75 | 85.34 | |
| Qwen3-VL-8BEvaluation Protocol=Fine-tuned2026.01 | 2.75 | 17.75 | 9.98 | 85.57 | |
| Llama3.2-11BEvaluation Protocol=Fine-tuned2026.01 | 2.17 | 15.12 | 9.09 | 85.15 | |
| Llama3.2-90BEvaluation Protocol=Pre-trained2026.01 | 2.09 | 12.84 | 8.05 | 84.76 | |
| gpt-5-nanoEvaluation Protocol=Pre-trained2026.01 | 1.85 | 11.56 | 6.33 | 84.37 | |
| LLaVA-MedEvaluation Protocol=Fine-tuned2026.01 | 1.56 | 10.72 | 6.04 | 84.67 | |
| Qwen2.5-VL-72BEvaluation Protocol=Pre-trained2026.01 | 1.48 | 17.28 | 9.97 | 82.11 | |
| LLaVA-NeXTEvaluation Protocol=Fine-tuned2026.01 | 1.48 | 11.25 | 6.07 | 84.45 |