Text-to-Image Generation on MS-COCO 2014 (FID-30K, CLIP-30K)
13.07FID (30K)Composer
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ComposerBackbone=SD2.1, Time=1.02 (+2%), Memory=8.4G2026.03 | 13.07 | 0.32 | |
| Test-time TrainingBackbone=SD2.1, Time=41.52 (+4,000%), Memory=55G2026.03 | 13.15 | 0.31 | |
| StandardBackbone=SD2.1, Time=1.00, Memory=8G2026.03 | 13.45 | 0.3 |