Image Generation on FFHQ 256x256 (test)
1.68FIDStyleSAN-XL
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| StyleSAN-XLNFE=1, ImageNet feature leakage=true2025.01 | 1.68 | — | — | — | — | — | — | — | |
| StyleNATDistillation=None2026.05 | 2.17 | — | — | — | — | 6.31 | 69.8 | 38.3 | |
| StyleGAN-XLparams (M)=67.9, Inference Time (sec/img)=0.0472023.03 | 2.19 | — | — | — | — | — | — | — | |
| StyleGAN-XLNFE=1, ImageNet feature leakage=true2025.01 | 2.19 | — | — | — | — | — | — | — | |
| StyleNAT+SDDistillation=SD2026.05 | 2.27 | — | — | — | — | 4.22 | 67.6 | 36.7 | |
| Poly-INRparams (M)=13.6, Inference Time (sec/img)=0.0542023.03 | 2.72 | — | — | — | — | — | — | — | |
| PolyINRNFE=1, ImageNet feature leakage=true2025.01 | 2.72 | — | — | — | — | — | — | — | |
| R3GAN (Ours-Config E)NFE=1, ImageNet feature leakage=false2025.01 | 2.75 | — | — | — | — | — | — | — | |
| ProjectedGAN+SDDistillation=SD2026.05 | 3.15 | — | — | — | — | 14.1 | 66.6 | 45.2 | |
| ProjectedGAN+SDL1Distillation=SDL12026.05 | 3.35 | — | — | — | — | 34.4 | 66.9 | 44.6 | |
| StyleGAN2-ADA+SDL1Distillation=SDL12026.05 | 3.46 | — | — | — | — | 4.51 | 67.9 | 46.9 | |
| StyleGAN2-ADA+SDDistillation=SD2026.05 | 3.5 | — | — | — | — | 1.18 | 67.8 | 44 | |
| StyleGAN2NFE=1, ImageNet feature leakage=false2025.01 | 3.78 | — | — | — | — | — | — | — | |
| StyleGAN2params (M)=30.0, Inference Time (sec/img)=0.0162023.03 | 3.83 | — | — | — | — | — | — | — | |
| ProjectedGANDistillation=None2026.05 | 3.9 | — | — | — | — | 79.3 | 66.2 | 43.9 | |
| StyleGAN3-RNFE=1, ImageNet feature leakage=false2025.01 | 3.92 | — | — | — | — | — | — | — | |
| StyleGAN2-ADADistillation=None2026.05 | 4 | — | — | — | — | 1.58 | 69.2 | 39.9 | |
| CIPSparams (M)=45.9, Inference Time (sec/img)=0.0672023.03 | 4.38 | — | — | — | — | — | — | — | |
| Original Full-Size StyleGAN2Hardware=V100 GPU, Img. Size=256, Backbone=StyleGAN22021.04 | 4.5 | 11.09 | — | — | — | — | — | — | |
| StyleGAN3-TNFE=1, ImageNet feature leakage=false2025.01 | 4.81 | — | — | — | — | — | — | — | |
| INR-GANparams (M)=72.4, Inference Time (sec/img)=0.0242023.03 | 4.95 | — | — | — | — | — | — | — | |
| LDMNFE=200, ImageNet feature leakage=false2025.01 | 4.98 | — | — | — | — | — | — | — | |
| Diffusion AutoencoderNFE=500, ImageNet feature leakage=false2025.01 | 5.81 | — | — | — | — | — | — | — | |
| Ours-CAHardware=V100 GPU, Img. Size=256, Backbone=StyleGAN22021.04 | 7.9 | 2.67 | — | — | — | — | — | — | |
| ADM (DPM-Solver)NFE=500, ImageNet feature leakage=false2025.01 | 8.4 | — | — | — | — | — | — | — | |
| ADM (DDIM)NFE=500, ImageNet feature leakage=false2025.01 | 8.41 | — | — | — | — | — | — | — | |
| LFRNFE=12026.01 | 8.75 | — | — | 0.0041 | 4.41 | — | — | — | |
| LFRNFE=102026.01 | 8.78 | — | — | 0.004 | 4.39 | — | — | — | |
| DFRNFE=102026.01 | 8.99 | — | — | 0.0048 | 4.51 | — | — | — | |
| DFRNFE=12026.01 | 9.04 | — | — | 0.0049 | 4.54 | — | — | — | |
| FMRefiner2026.01 | 9.27 | — | — | 0.0047 | 4.48 | — | — | — | |
| DiffuseVAE2026.01 | 9.32 | — | — | 0.0054 | 4.48 | — | — | — | |
| Base2026.01 | 9.47 | — | — | 0.0049 | 4.44 | — | — | — | |
| GSHardware=V100 GPU, Img. Size=256, Backbone=StyleGAN2, Implementation=Our Impl.2021.04 | 12.4 | 2.81 | — | — | — | — | — | — | |
| Full AttentionBackbone=DiT-S, Training Epochs=102025.12 | 38.77 | — | 61.64 | — | — | — | — | — | |
| LLSABackbone=DiT-S, Training Epochs=10, Top-K value=82025.12 | 39.29 | — | 375.34 | — | — | — | — | — | |
| SLABackbone=DiT-S, Training Epochs=10, Top-K value=322025.12 | 39.98 | — | 304.85 | — | — | — | — | — | |
| VSABackbone=DiT-S, Training Epochs=10, Top-K value=322025.12 | 40.69 | — | 341.94 | — | — | — | — | — |