Unconditional Image Generation on CelebA unconditional 64 x 64
1.21FIDDiffusion StyleGAN2+CS
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Diffusion StyleGAN2+CSBackbone=StyleGAN2023.12 | 1.21 | — | |
| PDM+CSBackbone=ADM, NFE=1000, Training=patch-wise2023.12 | 1.38 | — | |
| PDM+CSBackbone=ADM, NFE=50, Training=patch-wise2023.12 | 1.42 | — | |
| PDM+CSBackbone=ADM, NFE=100, Training=patch-wise2023.12 | 1.44 | — | |
| Diffusion StyleGAN2Backbone=StyleGAN2023.12 | 1.69 | — | |
| PDMBackbone=ADM, NFE=50, Training=patch-wise2023.12 | 1.77 | — | |
| PDMBackbone=ADM, NFE=1000, Training=patch-wise2023.12 | 1.82 | — | |
| PDMBackbone=ADM, NFE=100, Training=patch-wise2023.12 | 1.86 | — | |
| Soft TruncationArchitecture Type=Diffusion based on U-Net, #Params=62M, Advanced training techniques=true2022.09 | 1.9 | — | |
| PDM+CSBackbone=ADM, NFE=20, Training=patch-wise2023.12 | 1.97 | — | |
| DDIM+CSBackbone=ADM, NFE=10002023.12 | 1.98 | — | |
| DPM-Solver+CSBackbone=ADM, NFE=10002023.12 | 2.04 | — | |
| LEGO-S-PR#Params=35M, FLOPs=0.2G, Refinement=Progressive Refinement2023.10 | 2.09 | — | |
| DDIM+CSBackbone=ADM, NFE=1002023.12 | 2.11 | — | |
| DPM-Solver+CSBackbone=ADM, NFE=1002023.12 | 2.12 | — | |
| LEGO-S-PG#Params=35M, FLOPs=0.2G, Refinement=Progressive Growth2023.10 | 2.17 | — | |
| DPM-Solver+CSBackbone=ADM, NFE=502023.12 | 2.22 | — | |
| DDIM+CSBackbone=ADM, NFE=502023.12 | 2.23 | — | |
| DPM-Solver+CSBackbone=ADM, NFE=202023.12 | 2.35 | — | |
| DiT-S/2#Params=33M, FLOPs=0.4G2023.10 | 2.52 | — | |
| F-PNDMBackbone=ADM, NFE=1002023.12 | 2.81 | — | |
| F-PNDMBackbone=ADM, NFE=10002023.12 | 2.86 | — | |
| U-ViT-S/4#Params=44M, FLOPs=1.2G2023.10 | 2.87 | — | |
| U-ViT-S/4Architecture Type=Diffusion based on ViT, #Params=44M2022.09 | 2.87 | — | |
| SN-DDIMBackbone=ADM, NFE=10002023.12 | 2.9 | — | |
| SN-DDIMBackbone=ADM, NFE=1002023.12 | 3.04 | — | |
| DPM-SolverBackbone=ADM, NFE=10002023.12 | 3.08 | — | |
| DPM-SolverBackbone=ADM, NFE=502023.12 | 3.1 | — | |
| DPM-SolverBackbone=ADM, NFE=1002023.12 | 3.11 | — | |
| Analytic-DDIMBackbone=ADM, NFE=10002023.12 | 3.13 | — | |
| DPM-SolverBackbone=ADM, NFE=202023.12 | 3.13 | — | |
| DDIMArchitecture Type=Diffusion based on U-Net, #Params=79M2022.09 | 3.26 | — | |
| F-PNDMBackbone=ADM, NFE=502023.12 | 3.34 | — | |
| StyleGAN2Version=Generic version trained with DiffAug + bCR2021.07 | 3.39 | 3.43 | |
| ViTGAN2021.07 | 3.74 | 3.21 | |
| SN-DDIMBackbone=ADM, NFE=502023.12 | 3.83 | — | |
| PreSSnoise schedule=cosine2025.12 | 3.97 | 12.48 | |
| Analytic-DDIMBackbone=ADM, NFE=1002023.12 | 4.29 | — | |
| DDPMnoise schedule=cosine2025.12 | 4.58 | 11.43 | |
| DDIMBackbone=ADM, NFE=10002023.12 | 4.88 | — | |
| PDM+CSBackbone=ADM, NFE=10, Training=patch-wise2023.12 | 4.93 | — | |
| Diffusion AutoencoderBackbone=ADM, NFE=10002023.12 | 4.97 | — | |
| DDIM+CSBackbone=ADM, NFE=202023.12 | 5.11 | — | |
| D2CBackbone=NVAE+U-Net, NFE=10002023.12 | 5.15 | — | |
| DPM-Solver+CSBackbone=ADM, NFE=102023.12 | 5.22 | — | |
| Diffusion AutoencoderBackbone=ADM, NFE=1002023.12 | 5.3 | — | |
| TALORAPrec. (W/A)=6/6, Sampling steps=100, h (size of LoRA Hub)=42025.05 | 5.36 | 2.66 | |
| TALORAPrec. (W/A)=6/6, Sampling steps=100, h (size of LoRA Hub)=22025.05 | 5.38 | 2.67 | |
| F-PNDMBackbone=ADM, NFE=202023.12 | 5.51 | — | |
| D2CBackbone=NVAE+U-Net, NFE=1002023.12 | 5.7 | — | |
| DPM-SolverBackbone=ADM, NFE=102023.12 | 5.83 | — | |
| Analytic-DDIMBackbone=ADM, NFE=502023.12 | 6.13 | — | |
| FPPrec. (W/A)=32/32, Sampling steps=1002025.05 | 6.49 | 2.61 | |
| DDIMBackbone=ADM, NFE=1002023.12 | 6.53 | — | |
| SN-DDIMBackbone=ADM, NFE=202023.12 | 6.77 | — | |
| D2CBackbone=NVAE+U-Net, NFE=502023.12 | 6.8 | — | |
| GANFormer2Backbone=CNN+Transformer2023.12 | 6.87 | — | |
| Diffusion AutoencoderBackbone=ADM, NFE=502023.12 | 7.05 | — | |
| Cold DiffusionBackbone=ADM, NFE=1002023.12 | 7.13 | — | |
| P2 DiffusionBackbone=ADM, NFE=10002023.12 | 7.22 | — | |
| TALORAPrec. (W/A)=4/4, Sampling steps=100, h (size of LoRA Hub)=22025.05 | 7.69 | 2.59 | |
| F-PNDMBackbone=ADM, NFE=102023.12 | 7.71 | — | |
| DDIM+CSBackbone=ADM, NFE=102023.12 | 7.8 | — | |
| Cold DiffusionBackbone=ADM, NFE=10002023.12 | 7.84 | — | |
| TALORAPrec. (W/A)=4/4, Sampling steps=100, h (size of LoRA Hub)=42025.05 | 7.84 | 2.6 | |
| PDMBackbone=ADM, NFE=20, Training=patch-wise2023.12 | 8.36 | — | |
| P2 DiffusionBackbone=ADM, NFE=1002023.12 | 9.04 | — | |
| DDIMBackbone=ADM, NFE=502023.12 | 9.17 | — | |
| Diffusion AutoencoderBackbone=ADM, NFE=202023.12 | 10.18 | — | |
| SN-DDIMBackbone=ADM, NFE=102023.12 | 10.2 | — | |
| Analytic-DDIMBackbone=ADM, NFE=202023.12 | 10.45 | — | |
| D2CBackbone=NVAE+U-Net, NFE=202023.12 | 11.46 | — | |
| P2 DiffusionBackbone=ADM, NFE=502023.12 | 11.96 | — | |
| VQGANBackbone=CNN+Transformer2023.12 | 12.7 | — | |
| Diffusion AutoencoderBackbone=ADM, NFE=102023.12 | 12.92 | — | |
| Cold DiffusionBackbone=ADM, NFE=502023.12 | 12.93 | — | |
| DDIMBackbone=ADM, NFE=202023.12 | 13.73 | — | |
| Analytic-DDIMBackbone=ADM, NFE=102023.12 | 15.62 | — | |
| P2 DiffusionBackbone=ADM, NFE=202023.12 | 16.11 | — | |
| ADP-DMPrec. (W/A)=6/6, Sampling steps=1002025.05 | 16.86 | 2.3 | |
| D2CBackbone=NVAE+U-Net, NFE=102023.12 | 17.32 | — | |
| DDIMBackbone=ADM, NFE=102023.12 | 17.33 | — | |
| U-Net GANBackbone=U-Net2023.12 | 19.31 | — | |
| Vanilla-ViTDiscriminator=Vanilla ViT discriminator2021.07 | 20.2 | 2.57 | |
| P2 DiffusionBackbone=ADM, NFE=102023.12 | 20.37 | — | |
| Cold DiffusionBackbone=ADM, NFE=202023.12 | 20.45 | — | |
| Q-DiffusionPrec. (W/A)=6/6, Sampling steps=1002025.05 | 23.37 | 2.16 | |
| Cold DiffusionBackbone=ADM, NFE=102023.12 | 32.81 | — | |
| PDMBackbone=ADM, NFE=10, Training=patch-wise2023.12 | 35.88 | — | |
| NDA-BigGAN (Stitching)Resolution=64x64, NDA Strategy=Stitching2021.02 | 37.17 | — | |
| NDA-BigGAN (Jigsaw)Resolution=64x64, NDA Strategy=Jigsaw2021.02 | 37.24 | — | |
| NDA-BigGAN (Cutout)Resolution=64x64, NDA Strategy=Cutout2021.02 | 37.39 | — | |
| NDA-BigGAN (Cutmix)Resolution=64x64, NDA Strategy=Cutmix2021.02 | 37.46 | — | |
| NDA-BigGAN (Mixup)Resolution=64x64, NDA Strategy=Mixup2021.02 | 37.51 | — | |
| BigGANResolution=64x64, NDA Strategy=None2021.02 | 38.14 | — |