Image Generation on CelebA-HQ
4.3FIDCSDI-GAN
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CSDI-GANProposed=True2026.05 | 4.3 | — | 0.3392 | |
| U-ViTBackbone=U-ViT2025.12 | 4.303 | — | — | |
| B.I. GAN2026.05 | 4.6 | — | 0.2628 | |
| I-StyleGAN (Tr)2026.05 | 4.8 | — | 0.2799 | |
| StyleGAN2-ADA2026.05 | 5 | — | — | |
| MoG-da netSampling steps=1002026.06 | 5.1 | — | — | |
| SF-DiVeQbits=9, lr=2.5e-5, batch size=82025.09 | 5.21 | — | — | |
| NaiveSampling steps=1002026.06 | 5.23 | — | — | |
| STEbits=10, lr=2.5e-5, batch size=82025.09 | 5.28 | — | — | |
| RTbits=12, lr=2.5e-5, batch size=82025.09 | 5.44 | — | — | |
| MoG-a netSampling steps=1002026.06 | 5.45 | — | — | |
| ST-GSbits=12, lr=2.5e-5, batch size=82025.09 | 5.47 | — | — | |
| ST-GSbits=10, lr=2.5e-5, batch size=82025.09 | 5.48 | — | — | |
| STEbits=9, lr=2.5e-5, batch size=82025.09 | 5.57 | — | — | |
| SF-DiVeQbits=10, lr=2.5e-5, batch size=82025.09 | 5.57 | — | — | |
| MoG-da netSampling steps=502026.06 | 5.73 | — | — | |
| DiVeQbits=8, lr=2.5e-5, batch size=82025.09 | 5.9 | — | — | |
| NaiveSampling steps=502026.06 | 5.97 | — | — | |
| SF-DiVeQbits=12, lr=2.5e-5, batch size=82025.09 | 6 | — | — | |
| CSDI-GANRegularizer=w/o Lorth2026.05 | 6 | — | 0.2467 | |
| MoG-a netSampling steps=502026.06 | 6.23 | — | — | |
| EMAbits=12, lr=2.5e-5, batch size=82025.09 | 6.24 | — | — | |
| SF-DiVeQbits=8, lr=2.5e-5, batch size=82025.09 | 6.24 | — | — | |
| EMAbits=9, lr=2.5e-5, batch size=82025.09 | 6.3 | — | — | |
| EMAbits=10, lr=2.5e-5, batch size=82025.09 | 6.32 | — | — | |
| DiVeQbits=10, lr=2.5e-5, batch size=82025.09 | 6.32 | — | — | |
| RTbits=10, lr=2.5e-5, batch size=82025.09 | 6.4 | — | — | |
| Transitional-cGAN2026.05 | 6.4 | — | — | |
| RTbits=10, lr=2.5e-4, batch size=322025.09 | 6.53 | — | — | |
| RTbits=12, lr=2.5e-4, batch size=322025.09 | 6.58 | — | — | |
| DDIM (Uncond)Sampling steps=1002026.06 | 6.6 | — | — | |
| STEbits=8, lr=2.5e-5, batch size=82025.09 | 6.64 | — | — | |
| SF-DiVeQbits=9, lr=2.5e-4, batch size=322025.09 | 6.66 | — | — | |
| MoG-da netSampling steps=252026.06 | 6.68 | — | — | |
| STEbits=12, lr=2.5e-5, batch size=82025.09 | 6.69 | — | — | |
| DiVeQbits=9, lr=2.5e-5, batch size=82025.09 | 6.69 | — | — | |
| ST-GSbits=9, lr=2.5e-5, batch size=82025.09 | 6.81 | — | — | |
| EMAbits=8, lr=2.5e-5, batch size=82025.09 | 6.86 | — | — | |
| P2 weightingType=Diff, Sampling Steps=5002022.04 | 6.91 | — | — | |
| EMAbits=10, lr=2.5e-4, batch size=322025.09 | 6.97 | — | — | |
| NaiveSampling steps=252026.06 | 7 | — | — | |
| SF-DiVeQbits=10, lr=2.5e-4, batch size=322025.09 | 7.02 | — | — | |
| LSGMType=VAE+Diff2022.04 | 7.22 | — | — | |
| STEbits=10, lr=2.5e-4, batch size=322025.09 | 7.34 | — | — | |
| SF-DiVeQbits=12, lr=2.5e-4, batch size=322025.09 | 7.4 | — | — | |
| MoG-a netSampling steps=252026.06 | 7.41 | — | — | |
| EMAbits=9, lr=2.5e-4, batch size=322025.09 | 7.42 | — | — | |
| STEbits=9, lr=2.5e-4, batch size=322025.09 | 7.54 | — | — | |
| RTbits=9, lr=2.5e-5, batch size=82025.09 | 7.55 | — | — | |
| DiVeQbits=10, lr=2.5e-4, batch size=322025.09 | 7.59 | — | — | |
| CCModel configuration=Opt-interv., Base generator=StyleGAN2, Train time (hrs)=62025.03 | 7.65 | — | — | |
| CB-AEModel configuration=Base model, Base generator=StyleGAN2, Train time (hrs)=142025.03 | 7.66 | — | — | |
| CCModel configuration=Base model, Base generator=StyleGAN2, Train time (hrs)=62025.03 | 7.66 | — | — | |
| CB-AEModel configuration=Opt-interv., Base generator=StyleGAN2, Train time (hrs)=142025.03 | 7.67 | — | — | |
| DiVeQbits=12, lr=2.5e-5, batch size=82025.09 | 7.69 | — | — | |
| DiVeQbits=9, lr=2.5e-4, batch size=322025.09 | 8.01 | — | — | |
| PGGANType=GAN2022.04 | 8.03 | — | — | |
| EMAbits=8, lr=2.5e-4, batch size=322025.09 | 8.42 | — | — | |
| DiVeQbits=8, lr=2.5e-4, batch size=322025.09 | 8.44 | — | — | |
| SF-DiVeQbits=8, lr=2.5e-4, batch size=322025.09 | 8.46 | — | — | |
| ST-GSbits=8, lr=2.5e-5, batch size=82025.09 | 8.47 | — | — | |
| DDPMBackbone=DDPM2025.12 | 8.985 | — | — | |
| CBGMModel configuration=Base model, Base generator=StyleGAN2, Train time (hrs)=502025.03 | 9 | — | — | |
| CBGMModel configuration=CB model, Base generator=StyleGAN2, Train time (hrs)=502025.03 | 9.1 | — | — | |
| LDMBackbone=LDM2025.12 | 9.269 | — | — | |
| RTbits=8, lr=2.5e-5, batch size=82025.09 | 9.32 | — | — | |
| RTbits=9, lr=2.5e-4, batch size=322025.09 | 9.33 | — | — | |
| DDIM (Uncond)Sampling steps=502026.06 | 9.33 | — | — | |
| EMAbits=12, lr=2.5e-4, batch size=322025.09 | 9.41 | — | — | |
| STEbits=12, lr=2.5e-4, batch size=322025.09 | 9.45 | — | — | |
| CB-AEModel configuration=CB model, Base generator=StyleGAN2, Train time (hrs)=142025.03 | 9.52 | — | — | |
| DiVeQbits=12, lr=2.5e-4, batch size=322025.09 | 9.54 | — | — | |
| CB-AEModel configuration=CB interv., Base generator=StyleGAN2, Train time (hrs)=142025.03 | 9.65 | — | — | |
| MoG-da netSampling steps=102026.06 | 10.61 | — | — | |
| VQGANType=GAN+AR2022.04 | 10.7 | — | — | |
| NaiveSampling steps=102026.06 | 10.76 | — | — | |
| Improved DDPMBackbone=Improved DDPM2025.12 | 10.793 | — | — | |
| MoG-a netSampling steps=102026.06 | 11.45 | — | — | |
| RTbits=8, lr=2.5e-4, batch size=322025.09 | 12.3 | — | — | |
| DDIM (Uncond)Sampling steps=252026.06 | 13.45 | — | — | |
| FPModel=LDM-4, Bitwidth (W/A)=32/32, Steps=200, Eta=0.0, Sampler=PLMS2023.05 | 16.72 | 15.97 | — | |
| ALAEType=GAN2022.04 | 19.21 | — | — | |
| VAEBMType=VAE+EM2022.04 | 20.38 | — | — | |
| DDIM (Uncond)Sampling steps=102026.06 | 20.54 | — | — | |
| PTQDModel=LDM-4, Bitwidth (W/A)=4/8, Steps=200, Eta=0.0, Sampler=PLMS2023.05 | 21.08 | 17.38 | — | |
| Q-DiffusionModel=LDM-4, Bitwidth (W/A)=4/8, Steps=200, Eta=0.0, Sampler=PLMS2023.05 | 24.31 | 22.11 | — | |
| NVAEType=VAE2022.04 | 29.76 | — | — | |
| ST-GSbits=9, lr=2.5e-4, batch size=322025.09 | 41.1 | — | — | |
| NSVQbits=12, lr=2.5e-5, batch size=82025.09 | 48.9 | — | — | |
| NSVQbits=12, lr=2.5e-4, batch size=322025.09 | 49.6 | — | — | |
| NSVQbits=10, lr=2.5e-5, batch size=82025.09 | 59.2 | — | — | |
| NSVQbits=10, lr=2.5e-4, batch size=322025.09 | 62.1 | — | — | |
| GLOWType=Flow2022.04 | 68.93 | — | — | |
| NSVQbits=9, lr=2.5e-4, batch size=322025.09 | 70.1 | — | — | |
| NSVQbits=9, lr=2.5e-5, batch size=82025.09 | 70.4 | — | — | |
| NSVQbits=8, lr=2.5e-4, batch size=322025.09 | 78.4 | — | — | |
| NSVQbits=8, lr=2.5e-5, batch size=82025.09 | 81.5 | — | — | |
| VAE w/ N2CE-LEBMnz (latent dimension)=512, M (noise magnitude)=100, K (num. of stages)=32024.05 | 95.66 | — | — | |
| VAE w/ N2CE-LEBMnz (latent dimension)=512, M (noise magnitude)=100, K (num. of stages)=12024.05 | 101.71 | — | — | |
| VAE w/ MLE-LEBMnz (latent dimension)=5122024.05 | 111.11 | — | — |