Text-to-Image Alignment on Whoops
80.2VQAScoreIR-guidance
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| IR-guidanceBackbone=SD3.02026.06 | 80.2 | 29.6 | |
| ProjectionBackbone=SD3.0, Strategy=Projection2026.06 | 79.2 | 29.3 | |
| ResidualBackbone=SD3.0, Strategy=Residual2026.06 | 77.9 | 29.1 | |
| BaselineBackbone=SD3.02026.06 | 77.8 | 29.1 | |
| NoneBackbone=SD3.0, Strategy=None2026.06 | 77.3 | 29.3 | |
| Mean-StdBackbone=SD3.0, Strategy=Mean-Std2026.06 | 76 | 29.1 | |
| IR-guidanceBackbone=SD2.12026.06 | 64.1 | 28.1 | |
| NoneBackbone=SD2.1, Strategy=None2026.06 | 63.2 | 28.2 | |
| BaselineBackbone=SD2.12026.06 | 62.9 | 28 | |
| ProjectionBackbone=SD2.1, Strategy=Projection2026.06 | 62.7 | 28 | |
| ResidualBackbone=SD2.1, Strategy=Residual2026.06 | 61.7 | 27.9 | |
| Mean-StdBackbone=SD2.1, Strategy=Mean-Std2026.06 | 61.7 | 27.8 |