Aesthetic Quality Assessment on UltraVQA
85Accuracy @0.5ASO
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ASOModel Category=Ours, Protocol=Analytic Score Optimization2026.02 | 85 | 0.357 | 0.824 | 0.837 | |
| GRPOModel Category=Ours, Protocol=Group Relative Policy Optimization2026.02 | 84.7 | 0.352 | 0.836 | 0.841 | |
| SFTModel Category=Ours, Protocol=Supervised Fine-Tuning2026.02 | 74 | 0.467 | 0.752 | 0.766 | |
| FineVQModel Category=VQA Baselines2026.02 | 72.5 | 0.885 | 0.743 | 0.731 | |
| VideoScoreV2Model Category=VQA Baselines2026.02 | 63.7 | 1.072 | 0.651 | 0.639 | |
| GPT-4.1Model Category=VLM-APIs2026.02 | 56.9 | 0.744 | 0.611 | 0.615 | |
| Gemini-2.5ProModel Category=VLM-APIs2026.02 | 56 | 0.812 | 0.654 | 0.643 | |
| VideoLLaMA3Model Category=Open-source VLMs2026.02 | 50.2 | 0.973 | 0.452 | 0.49 | |
| InternVideo-2.5Model Category=Open-source VLMs2026.02 | 45.7 | 1.026 | 0.308 | 0.296 | |
| MiniCPMV-4.5Model Category=Open-source VLMs2026.02 | 44.3 | 1.052 | 0.389 | 0.394 | |
| Q-AlignModel Category=VQA Baselines2026.02 | 44.3 | 0.975 | 0.616 | 0.602 | |
| VideoChat-FlashModel Category=Open-source VLMs2026.02 | 43.5 | 1.121 | 0.206 | 0.124 | |
| Qwen2.5-VLModel Category=Open-source VLMs2026.02 | 41 | 1.187 | 0.373 | 0.361 | |
| InternVL-3.5Model Category=Open-source VLMs2026.02 | 40.3 | 1.192 | 0.508 | 0.505 |