Image Generation on ImageNet 256x256 1k (test val)
1.93FIDSA-Solver
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SA-SolverArchitecture=Min-SNR DiT, NFE=20, Classifier-free guidance scale (s)=1.52023.09 | 1.93 | — | — | — | |
| SA-SolverArchitecture=DiT, NFE=60, Classifier-free guidance scale (s)=1.52023.09 | 2.02 | — | — | — | |
| HeunArchitecture=Min-SNR DiT, NFE=50, Classifier-free guidance scale (s)=1.52023.09 | 2.06 | — | — | — | |
| DDPMArchitecture=DiT, NFE=250, Classifier-free guidance scale (s)=1.52023.09 | 2.27 | — | — | — | |
| VRPOTraining Speed=2.3×2026.05 | 4.12 | 302.7 | 78.1 | 76 | |
| REGTraining Speed=1.5×2026.05 | 5.74 | 289.1 | 76.2 | 72 | |
| REPATraining Speed=1.4×2026.05 | 5.93 | 285.6 | 75.5 | 70 | |
| SiTTraining Speed=1.2×2026.05 | 6.84 | 274.5 | 73.8 | 64 | |
| DiTTraining Speed=1.0×2026.05 | 7.95 | 260.3 | 72.4 | 61 |