Unconditional Image Generation on CIFAR-10
1.36FIDSAN
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SANTraining Regime=Training From Scratch, NFE=12024.05 | 1.36 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SiD + N^2CE (M=50)Training Regime=Post-training, NFE=12024.05 | 1.45 | — | — | — | 10.23 | — | — | — | — | — | — | — | — | |
| CAF + GANN=12024.11 | 1.48 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SiD2AFamily=Diffusion & GAN, NFE=1, αSiD=1.02025.09 | 1.5 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SiD^2ATraining Regime=Post-training, NFE=12024.05 | 1.5 | — | — | — | 10.19 | — | — | — | — | — | — | — | — | |
| SiDAFamily=Diffusion & GAN, NFE=1, αSiD=1.02025.09 | 1.52 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SiD2AFamily=Diffusion & GAN, NFE=1, αSiD=1.22025.09 | 1.52 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SiD + NCE (M=1)Training Regime=Post-training, NFE=12024.05 | 1.53 | — | — | — | 10.2 | — | — | — | — | — | — | — | — | |
| SADMDiffusion Space=Data, NFE=352024.02 | 1.54 | — | — | — | — | — | 2.28 | — | — | — | — | — | — | |
| D2O-FTraining Regime=Post-training, NFE=12024.05 | 1.54 | — | — | — | 10.1 | — | — | — | — | — | — | — | — | |
| EDM-G++Diffusion Space=Data, NFE=352024.02 | 1.77 | — | — | — | — | — | 2.55 | — | — | — | — | — | — | |
| SADMDiffusion Space=Latent, NFE=1382024.02 | 1.78 | — | — | — | — | — | 2.51 | — | — | — | — | — | — | |
| StyleGAN-XLNFE=12024.06 | 1.85 | — | — | — | — | — | — | — | 0.47 | — | — | — | — | |
| CTMNFE=2, Category=Distillation Sampling2023.10 | 1.87 | — | — | — | — | — | 2.43 | — | — | — | — | — | — | |
| CTMN=22024.11 | 1.87 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CTMTraining Regime=Post-training, NFE=22024.05 | 1.87 | — | — | — | — | — | — | — | — | — | — | — | — | |
| RestartModel=PFGM++, NFE=432023.06 | 1.88 | — | — | — | — | — | — | — | — | — | — | — | — | |
| RestartModel=EDM-VP, NFE=432023.06 | 1.9 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ODE (Heun)Model=PFGM++, NFE=632023.06 | 1.91 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ODE (Heun)Model=PFGM++, NFE=352023.06 | 1.91 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SiDFamily=Diffusion & GAN, NFE=1, αSiD=1.22025.09 | 1.92 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SiDTraining Regime=Post-training, NFE=12024.05 | 1.92 | — | — | — | 9.98 | — | — | — | — | — | — | — | — | |
| LSGM-G++Diffusion Space=Latent, NFE=1382024.02 | 1.94 | — | — | — | — | — | 3.42 | — | — | — | — | — | — | |
| Heun + FDSNFE=199, Denoising steps=1002026.04 | 1.954 | — | — | — | 9.633 | — | — | — | — | — | — | — | — | |
| EDMArchitecture Type=Diffusion based on U-Net, #Params=56M, Advanced training techniques=true2022.09 | 1.97 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ODE (Heun)Model=EDM-VP, NFE=632023.06 | 1.97 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ODE (Heun)Model=EDM-VP, NFE=352023.06 | 1.97 | — | — | — | — | — | — | — | — | — | — | — | — | |
| EDMDiffusion Space=Data, NFE=352024.02 | 1.97 | — | — | — | — | — | 2.6 | — | — | — | — | — | — | |
| VP-EDMFamily=Diffusion & GAN, NFE=352025.09 | 1.97 | — | — | — | — | — | — | — | — | — | — | — | — | |
| EDMModel group=Frequency-agnostic model2026.05 | 1.97 | — | — | — | 9.89 | — | — | — | — | — | — | — | — | |
| CTMNFE=1, Category=Distillation Sampling2023.10 | 1.98 | — | — | — | — | — | 2.43 | — | — | — | — | — | — | |
| CTMN=12024.11 | 1.98 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CTMFamily=Diffusion & GAN, NFE=12025.09 | 1.98 | — | — | — | — | — | — | — | — | — | — | — | — | |
| RealUIDFamily=Flow-based, NFE=1, Fine-tuned=true2025.09 | 1.98 | — | — | — | — | — | — | — | — | — | — | — | — | |
| EDMTraining Regime=Training From Scratch, NFE=352024.05 | 1.98 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CTMTraining Regime=Post-training, NFE=12024.05 | 1.98 | — | — | — | — | — | — | — | — | — | — | — | — | |
| EDMNFE=352023.10 | 2.01 | — | — | — | — | — | 2.56 | — | — | — | — | — | — | |
| EDMN=352024.11 | 2.01 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SiDFamily=Diffusion & GAN, NFE=1, αSiD=1.02025.09 | 2.03 | — | — | — | — | — | — | — | — | — | — | — | — | |
| HeunNFE=199, Denoising steps=1002026.04 | 2.038 | — | — | — | 9.606 | — | — | — | — | — | — | — | — | |
| EDMNFE=352025.09 | 2.04 | — | — | — | 9.84 | — | — | — | — | — | — | — | — | |
| Heun†NFE=307, Denoising steps=1002026.04 | 2.042 | — | — | — | 9.603 | — | — | — | — | — | — | — | — | |
| Euler + FDSNFE=100, Denoising steps=1002026.04 | 2.056 | — | — | — | 9.519 | — | — | — | — | — | — | — | — | |
| LSGM2022.05 | 2.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LSGMDiffusion Space=Latent, NFE=1382024.02 | 2.1 | — | — | — | — | — | 3.43 | — | — | — | — | — | — | |
| LSGMNFE=1382023.10 | 2.1 | — | — | — | — | — | 3.43 | — | — | — | — | — | — | |
| LSGMN=1382024.11 | 2.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LSGMNFE=1472025.09 | 2.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LowDiffNFE=22†2025.09 | 2.11 | — | — | — | 9.87 | — | — | — | — | — | — | — | — | |
| Subspace Diffusionvariant=subspace, backbone=NCSN++ (deep), divergence threshold=1e-3, downsampling time t1=0.562022.05 | 2.17 | — | — | — | 9.94 | 0.69 | — | — | — | — | — | — | — | |
| BiFMSetting=Multi-Step, NFE=502026.03 | 2.17 | — | — | — | — | — | — | — | — | — | — | — | — | |
| NCSN++ (deep)variant=full, backbone=NCSN++ (deep)2022.05 | 2.2 | — | — | — | 9.89 | — | — | — | — | — | — | — | — | |
| NCSN++Diffusion Space=Data, NFE=20002024.02 | 2.2 | — | — | — | — | — | 3.45 | — | — | — | — | — | — | |
| Score SDENFE=20002023.10 | 2.2 | — | — | — | — | — | 3.45 | — | — | — | — | — | — | |
| Score SDEN=20002024.11 | 2.2 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Score SDE (VE)NFE=20002024.06 | 2.2 | — | — | — | — | — | — | — | 0.59 | — | — | — | — | |
| Score SDENFE=20002025.09 | 2.2 | — | — | — | 9.89 | — | — | — | — | — | — | — | — | |
| NCSN++Model group=Frequency-agnostic model2026.05 | 2.2 | — | — | — | 9.64 | — | — | — | — | — | — | — | — | |
| CLD-SGM2022.05 | 2.23 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Euler†NFE=154, Denoising steps=1002026.04 | 2.238 | — | — | — | 9.543 | — | — | — | — | — | — | — | — | |
| iCT-deepTraining Regime=Training From Scratch, NFE=22024.05 | 2.24 | — | — | — | 9.89 | — | — | — | — | — | — | — | — | |
| CLD-SGMDiffusion Space=Data, NFE=3122024.02 | 2.25 | — | — | — | — | — | 3.31 | — | — | — | — | — | — | |
| Improved SDEModel=EDM-VP, NFE=632023.06 | 2.27 | — | — | — | — | — | — | — | — | — | — | — | — | |
| INDMDiffusion Space=Latent, NFE=20002024.02 | 2.28 | — | — | — | — | — | 3.09 | — | — | — | — | — | — | |
| Subspace Diffusionvariant=subspace, backbone=NCSN++, divergence threshold=3e-3, downsampling time t1=0.522022.05 | 2.29 | — | — | — | 9.99 | 0.66 | — | — | — | — | — | — | — | |
| PFGMNFE=1102025.09 | 2.35 | — | — | — | 9.68 | — | — | — | — | — | — | — | — | |
| Bilateral DPM2022.05 | 2.38 | — | — | — | — | — | — | — | — | — | — | — | — | |
| NCSN++variant=full, backbone=NCSN++2022.05 | 2.38 | — | — | — | 9.83 | — | — | — | — | — | — | — | — | |
| CTMNFE=1, Category=Direct Generation2023.10 | 2.39 | — | — | — | — | — | — | — | — | — | — | — | — | |
| EulerNFE=100, Denoising steps=1002026.04 | 2.396 | — | — | — | 9.493 | — | — | — | — | — | — | — | — | |
| Subspace Diffusionvariant=subspace, backbone=DDPM++ (deep), divergence threshold=1e-4, downsampling time t1=0.562022.05 | 2.4 | — | — | — | 9.66 | 0.69 | — | — | — | — | — | — | — | |
| DDPM++ (deep)variant=full, backbone=DDPM++ (deep)2022.05 | 2.41 | — | — | — | 9.57 | — | — | — | — | — | — | — | — | |
| Adaptive solver2022.05 | 2.44 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Improved SDEModel=EDM-VP, NFE=352023.06 | 2.45 | — | — | — | — | — | — | — | — | — | — | — | — | |
| iCTTraining Regime=Training From Scratch, NFE=22024.05 | 2.46 | — | — | — | 9.8 | — | — | — | — | — | — | — | — | |
| Soft TruncationDiffusion Space=Data, NFE=20002024.02 | 2.47 | — | — | — | — | — | 2.91 | — | — | — | — | — | — | |
| iCT-deepFamily=Diffusion & GAN, NFE=12025.09 | 2.51 | — | — | — | — | — | — | — | — | — | — | — | — | |
| iCT-deepTraining Regime=Training From Scratch, NFE=12024.05 | 2.51 | — | — | — | 9.76 | — | — | — | — | — | — | — | — | |
| COT-FMBaseline framework=MeanFlow, Sampling steps=2-step2026.03 | 2.53 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Diffusion ProjectedGANFamily=Diffusion & GAN, NFE=12025.09 | 2.54 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DDPM++ cont.Architecture Type=Diffusion based on U-Net, #Params=62M2022.09 | 2.55 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Prog. Distillation2022.05 | 2.57 | — | — | — | — | — | — | — | — | — | — | — | — | |
| PDNFE=82024.06 | 2.57 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Subspace Diffusionvariant=subspace, backbone=DDPM++, divergence threshold=3e-5, downsampling time t1=0.622022.05 | 2.6 | — | — | — | 9.54 | 0.73 | — | — | — | — | — | — | — | |
| COT-FMBaseline framework=MeanFlow, Sampling steps=1-step2026.03 | 2.6 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DDPM++variant=full, backbone=DDPM++2022.05 | 2.61 | — | — | — | 9.56 | — | — | — | — | — | — | — | — | |
| DMDTraining Regime=Post-training, NFE=12024.05 | 2.62 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Flow MatchingSetting=Multi-Step, NFE=502026.03 | 2.63 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CD + GANNFE=12023.10 | 2.65 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SKILDModel group=Frequency/scale-informed model2026.05 | 2.65 | — | — | — | 9.63 | — | — | — | — | — | — | — | — | |
| FACMFamily=Flow-based, NFE=12025.09 | 2.69 | — | — | — | — | — | — | — | — | — | — | — | — | |
| StyleGAN2+ADAFamily=Diffusion & GAN, NFE=1, Fine-tuned=true, DI=true2025.09 | 2.71 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Complexity-Balanced SplittingArchitecture=UNet, Split Boundaries=0.2, 0.782026.06 | 2.72 | — | — | — | — | — | — | — | — | — | — | — | — | |
| BiFMSetting=One-Step, NFE=12026.03 | 2.75 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DDPM++Model group=Frequency-agnostic model2026.05 | 2.78 | — | — | — | 9.46 | — | — | — | — | — | — | — | — | |
| iCTNFE=12025.12 | 2.83 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ESCNFE=12025.12 | 2.83 | — | — | — | — | — | — | — | — | — | — | — | — | |
| iCTTraining Regime=Training From Scratch, NFE=12024.05 | 2.83 | — | — | — | 9.54 | — | — | — | — | — | — | — | — | |
| sCTFamily=Diffusion & GAN, NFE=12025.09 | 2.85 | — | — | — | — | — | — | — | — | — | — | — | — | |
| sCTSetting=One-Step, NFE=12026.03 | 2.85 | — | — | — | — | — | — | — | — | — | — | — | — |