Text-to-Image Generation on Whoops (CLIP, VQA, CHAIR)
0.296CLIP ScoreIR-guidance
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| IR-guidancebackbone=SD3.02026.06 | 0.296 | 80.2 | 61.1 | 95 | 47.6 | |
| Stable Diffusion 3.0variant=baseline2026.06 | 0.291 | 77.8 | 61.6 | 96 | 45.6 | |
| IR-guidancebackbone=SD2.12026.06 | 0.281 | 64.1 | 68.1 | 97.2 | 38.4 | |
| Stable Diffusion 2.1variant=baseline2026.06 | 0.28 | 62.9 | 66.4 | 97.8 | 39.7 |