Image Styling on Regular Text
78.77Overall ScoreStandardized Reward-Weighted (SW)
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Standardized Reward-Weighted (SW)Model Size=4B2026.03 | 78.77 | 76.58 | 82.84 | 81.55 | 80.19 | 77.55 | 73.94 | |
| Reward-Weighted (RW)Model Size=4B2026.03 | 77.18 | 73.61 | 81.35 | 80 | 78.84 | 76.84 | 72.45 | |
| Reward (R)Model Size=4B2026.03 | 77.12 | 72.84 | 83.03 | 81.42 | 79.61 | 75.16 | 70.71 | |
| GPT-4o (G4o)Strategy=zero-shot baseline2026.03 | 76.07 | 75.61 | 76.39 | 76.76 | 74.13 | 73.42 | 72.71 | |
| Baseline (B)Model Size=4B2026.03 | 76.03 | 72.58 | 80.9 | 80 | 77.68 | 74.65 | 70.45 | |
| SFT (S)Model Size=4B2026.03 | 75.03 | 70.65 | 81.16 | 80.19 | 77.35 | 73.35 | 67.42 | |
| DPO (D)Model Size=4B2026.03 | 74.88 | 71.03 | 79.81 | 79.48 | 76.32 | 72.65 | 69.94 | |
| Edit-Only (E)Model Size=4B2026.03 | 71.49 | — | 75.81 | — | — | 68.13 | — |