Super-Resolution on HGGT (test-100)
72.69MUSIQ ScoreRESRGAN-UPos + FT
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| RESRGAN-UPos + FTBackbone=Real-ESRGAN, No human GT ranking data=false, Protocol=Direct fine-tuning2025.04 | 72.69 | 21.09 | 0.58 | 0.235 | 0.179 | 0.163 | 5.37 | 3.69 | 0.71 | |
| RESRGAN-UPos + FTIGBackbone=Real-ESRGAN, No human GT ranking data=false, Protocol=Fine-tuning with GAN loss2025.04 | 72.01 | 21.14 | 0.586 | 0.236 | 0.182 | 0.16 | 5.35 | 3.7 | 0.7 | |
| RESRGAN-AMO + FTBackbone=Real-ESRGAN, No human GT ranking data=true, Protocol=Direct fine-tuning2025.04 | 71.67 | 21.02 | 0.581 | 0.228 | 0.169 | 0.161 | 5.35 | 3.68 | 0.71 | |
| SwinIR-AMO + FTBackbone=SwinIR, No human GT ranking data=true, Protocol=Direct fine-tuning2025.04 | 70.81 | 21.77 | 0.624 | 0.174 | 0.121 | 0.128 | 5.29 | 3.75 | 0.7 | |
| RESRGAN-UPos + FTHPBackbone=Real-ESRGAN, No human GT ranking data=false, Protocol=Fine-tuning with higher perceptual loss weight2025.04 | 70.28 | 21.3 | 0.595 | 0.226 | 0.175 | 0.158 | 5.32 | 3.65 | 0.69 | |
| SwinIR-UPos + FTBackbone=SwinIR, No human GT ranking data=false, Protocol=Direct fine-tuning2025.04 | 69.7 | 22.01 | 0.633 | 0.169 | 0.123 | 0.124 | 5.26 | 3.7 | 0.67 | |
| Gold StandardNo human GT ranking data=false2025.04 | 69.64 | — | — | — | — | — | 5.28 | 3.78 | 0.69 | |
| SwinIR-UPos + FTIGBackbone=SwinIR, No human GT ranking data=false, Protocol=Fine-tuning with GAN loss2025.04 | 69.37 | 22.03 | 0.635 | 0.168 | 0.122 | 0.123 | 5.24 | 3.69 | 0.66 | |
| SwinIR-UPos + FTHPBackbone=SwinIR, No human GT ranking data=false, Protocol=Fine-tuning with higher perceptual loss weight2025.04 | 68.38 | 22.17 | 0.642 | 0.166 | 0.123 | 0.122 | 5.23 | 3.64 | 0.65 | |
| SwinIR-AMOBackbone=SwinIR, No human GT ranking data=true2025.04 | 68.08 | 22.08 | 0.641 | 0.167 | 0.124 | 0.123 | 5.21 | 3.67 | 0.66 | |
| RESRGAN-AMOBackbone=Real-ESRGAN, No human GT ranking data=true2025.04 | 67.86 | 21.28 | 0.602 | 0.224 | 0.178 | 0.156 | 5.29 | 3.56 | 0.66 | |
| SwinIR-SMPBackbone=SwinIR, No human GT ranking data=false2025.04 | 66.83 | 22.29 | 0.647 | 0.171 | 0.13 | 0.124 | 5.17 | 3.62 | 0.62 | |
| SwinIR-SMABackbone=SwinIR, No human GT ranking data=true2025.04 | 66.73 | 22.27 | 0.646 | 0.171 | 0.129 | 0.124 | 5.16 | 3.6 | 0.63 | |
| RESRGAN-SMPBackbone=Real-ESRGAN, No human GT ranking data=false2025.04 | 66.66 | 21.44 | 0.607 | 0.226 | 0.182 | 0.156 | 5.24 | 3.51 | 0.64 | |
| SwinIR-UPos*Backbone=SwinIR, No human GT ranking data=false, Sampling=Uniform sampling from human-ranked positives2025.04 | 66.39 | 22.3 | 0.647 | 0.169 | 0.129 | 0.123 | 5.16 | 3.56 | 0.62 | |
| RESRGAN-UPos*Backbone=Real-ESRGAN, No human GT ranking data=false2025.04 | 65.93 | 21.54 | 0.608 | 0.233 | 0.192 | 0.158 | 5.25 | 3.47 | 0.63 | |
| RESRGAN-SMABackbone=Real-ESRGAN, No human GT ranking data=true2025.04 | 65.87 | 21.46 | 0.606 | 0.227 | 0.182 | 0.157 | 5.23 | 3.46 | 0.63 | |
| SwinIR-RandBackbone=SwinIR, No human GT ranking data=true, Sampling=Random GT choice2025.04 | 65.27 | 22.45 | 0.65 | 0.18 | 0.139 | 0.131 | 5.11 | 3.52 | 0.59 | |
| RESRGAN-RandBackbone=Real-ESRGAN, No human GT ranking data=true2025.04 | 64.82 | 21.66 | 0.611 | 0.234 | 0.19 | 0.16 | 5.18 | 3.4 | 0.6 | |
| SwinIR-OrigsOnlyBackbone=SwinIR, No human GT ranking data=true, Supervision=No enhanced GT2025.04 | 59.47 | 22.72 | 0.652 | 0.227 | 0.174 | 0.162 | 4.87 | 3.17 | 0.48 | |
| RESRGAN-OrigsOnlyBackbone=Real-ESRGAN, No human GT ranking data=true2025.04 | 57.91 | 22.1 | 0.618 | 0.283 | 0.229 | 0.185 | 4.84 | 2.99 | 0.46 |