Fine-Grained Aesthetic Assessment (Pair-level) on FGAesthetics Cropping
77.4AccuracyFGAesQ (w DiffToken)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FGAesQ (w DiffToken)Params=86.3M, DiffToken=Enabled2026.03 | 77.4 | 77.3 | |
| MUSIQ (Fine-tuned)Params=78.6M, Evaluation Protocol=Fine-tuned2026.03 | 77 | 77 | |
| FGAesQ (w/o DiffToken)Params=86.3M, DiffToken=Disabled2026.03 | 76.4 | 76.3 | |
| Charm (Fine-tuned)Params=85.7M, Evaluation Protocol=Fine-tuned2026.03 | 75.5 | 75.5 | |
| Q-Align (ICML 2024)Params=8.20B2026.03 | 73.8 | 73.9 | |
| MUSIQ (ICCV 2021)Params=78.6M2026.03 | 73.1 | 73.1 | |
| VILA (CVPR 2023)Params=303M2026.03 | 72.6 | 72.6 | |
| VILA (Fine-tuned)Params=303M, Evaluation Protocol=Fine-tuned2026.03 | 71.4 | 71 | |
| MLSP (CVPR 2019)Params=114 M2026.03 | 71 | 71 | |
| NIMA (Fine-tuned)Params=54.3M, Evaluation Protocol=Fine-tuned2026.03 | 70.8 | 70.4 | |
| Charm (CVPR 2025)Params=85.7M2026.03 | 70.7 | 70.7 | |
| UNIAA (arXiv 2024)Params=7.06B2026.03 | 67.5 | 67 | |
| NIMA (TIP 2018)Params=54.3M2026.03 | 65.5 | 64.9 | |
| TANet (IJCAI 2022)Params=13.9M2026.03 | 64.1 | 64.1 | |
| RealQA (arXiv 2025)Params=8.29B2026.03 | 60.5 | 61.3 |