Quality Improving on RETOUCHEVAL 1.0 (test)
26.41L1RETOUCHIQ
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RETOUCHIQStage=SFT2026.02 | 26.41 | 42.1 | 5.98 | 6.84 | 6.4 | |
| RETOUCHIQStage=RL (Generalist Reward Model)2026.02 | 27.03 | 42.73 | 6.39 | 6.92 | 6.65 | |
| RETOUCHIQStage=RL (Rule-based Reward)2026.02 | 28.22 | 43.89 | 6.2 | 6.83 | 6.51 | |
| MONETGPT2026.02 | 30.08 | 44.98 | 6.27 | 6.7 | 6.48 | |
| RETOUCHIQStage=RL (Generalist Reward Model)2026.02 | 31.41 | 44.99 | 7.57 | 7.48 | 7.51 | |
| MONETGPT2026.02 | 32.05 | 44.98 | 6.52 | 7.05 | 6.78 | |
| JARVISART2026.02 | 32.17 | 46.23 | 7.22 | 6.59 | 6.9 | |
| RETOUCHIQStage=RL (Rule-based Reward)2026.02 | 32.65 | 46.47 | 7.14 | 6.61 | 6.87 | |
| RETOUCHIQStage=SFT2026.02 | 35.03 | 46.63 | 6.71 | 6.67 | 6.69 | |
| GEMINI-2.52026.02 | 36.11 | 46.72 | 6.01 | 6.63 | 6.31 | |
| GPT-52026.02 | 36.21 | 47.53 | 6.74 | 6.51 | 6.62 | |
| GPT-52026.02 | 37.2 | 47.56 | 6.07 | 6.89 | 6.47 | |
| JARVISART2026.02 | 38.19 | 49.75 | 6.45 | 6.39 | 6.42 | |
| GEMINI-2.52026.02 | 39.58 | 51.6 | 6.56 | 6.72 | 6.64 | |
| FLUX-PRO2026.02 | 64.97 | 98.14 | 6.12 | 6.09 | 6.1 | |
| FLUX-PRO2026.02 | 67.93 | 82.86 | 6.19 | 6.26 | 6.22 |