Image Generation on ImageNet 256x256 50k (test)
1.42FID (ImageNet 50k)SiT-XL/2+REPA
Evaluation Results
| Method | Links | |
|---|---|---|
| SiT-XL/2+REPAFamily=Diff/ Flow, Param.=675M, NFE=250x22025.12 | 1.42 | |
| MAR-HFamily=AR/Mask, Param.=943M, NFE=256x22025.12 | 1.55 | |
| VAR-d30Family=AR/Mask, Param.=2B, NFE=10x22025.12 | 1.92 | |
| SiT-XL/2Family=Diff/ Flow, Param.=675M, NFE=250x22025.12 | 2.06 | |
| DiT-XL/2Family=Diff/ Flow, Param.=675M, NFE=250x22025.12 | 2.27 | |
| StyleGAN-XLFamily=GAN, Param.=166M, NFE=12025.12 | 2.3 | |
| ESC+ (w/-class-consist.)Family=Shortcut, Param.=676M, NFE=12025.12 | 2.53 | |
| SimDiffFamily=Diff/ Flow, Param.=2B, NFE=512x22025.12 | 2.77 | |
| ESC (w/-class-consist.)Family=Shortcut, Param.=676M, NFE=12025.12 | 2.85 | |
| ESC (w/o-class-consist.)Family=Shortcut, Param.=676M, NFE=12025.12 | 2.92 | |
| MeanFlowFamily=Shortcut, Param.=676M, NFE=22025.12 | 2.93 | |
| MeanFlowFamily=Shortcut, Param.=676M, NFE=12025.12 | 3.43 | |
| GigaGANFamily=GAN, Param.=569M, NFE=12025.12 | 3.45 | |
| LDM-4-GFamily=Diff/ Flow, Param.=400M, NFE=250x22025.12 | 3.6 | |
| MaskGITFamily=AR/Mask, Param.=227M, NFE=82025.12 | 6.18 | |
| BigGANFamily=GAN, Param.=112M, NFE=12025.12 | 6.95 | |
| IMMFamily=Shortcut, Param.=675M, NFE=1x22025.12 | 7.77 | |
| SCDFamily=Shortcut, Param.=675M, NFE=12025.12 | 10.6 | |
| ADMFamily=Diff/ Flow, Param.=554M, NFE=250x22025.12 | 10.94 | |
| AR w/ VQGANFamily=AR/Mask, Param.=227M, NFE=10242025.12 | 26.52 | |
| iCTFamily=Shortcut, Param.=675M, NFE=12025.12 | 34.24 |