Image Generation on MNIST
0.52FIDSVDM
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| SVDMH=0.9, N=22026.06 | 0.52 | — | — | — | — | — | — | — | — | |
| Real2026.01 | 1.02 | — | — | — | 9.8 | — | — | — | — | |
| LNDSM-SGM2025.12 | 2.2 | — | — | — | 9.75 | — | — | — | — | |
| cU-Net with DDIMsampler=DDIM2026.06 | 2.55 | — | — | — | — | — | — | — | — | |
| Normalizing Flow PreconditionerPreconditioner=Normalizing Flow2026.03 | 2.62 | — | — | — | — | — | — | — | — | |
| SIGCategory=iterative, T=12020.07 | 4.5 | — | — | — | — | — | — | — | — | |
| VPGA2026.06 | 4.97 | — | — | — | — | — | — | — | — | |
| GANetic-loss DCGAN2026.06 | 5.96 | — | — | — | — | — | — | — | — | |
| WGANCategory=adversarial training2020.07 | 6.7 | — | — | — | — | — | — | — | — | |
| LSGM2025.12 | 6.7 | — | — | — | 9.42 | — | — | — | — | |
| Flow Matching PreconditionerPreconditioner=Flow Matching2026.03 | 6.95 | — | — | — | — | — | — | — | — | |
| GFDM2026.06 | 7.34 | — | — | — | — | — | — | — | — | |
| Brownian SDE baseline2026.06 | 7.49 | — | — | — | — | — | — | — | — | |
| PrivImageepsilon=12026.01 | 7.6 | — | — | — | — | 94 | — | — | — | |
| DuelGANreported_in_original_paper=false2021.01 | 7.87 | — | — | — | — | — | — | — | — | |
| FETA-Proepsilon=12026.01 | 8 | — | — | — | — | 97.3 | — | — | — | |
| PDP-Diffusionepsilon=12026.01 | 8.9 | — | — | — | — | 94.5 | — | — | — | |
| PrivImage-Proepsilon=12026.01 | 9.9 | — | — | — | — | 94.7 | — | — | — | |
| Best default GANCategory=adversarial training2020.07 | 10 | — | — | — | — | — | — | — | — | |
| PDP-Diffusion-Proepsilon=12026.01 | 10.9 | — | — | — | — | 95.3 | — | — | — | |
| two-stage VAECategory=AE based2020.07 | 12.6 | — | — | — | — | — | — | — | — | |
| Flow MatchingPreconditioner=None2026.03 | 13.83 | — | — | — | — | — | — | — | — | |
| WGANreported_in_original_paper=true2021.01 | 14.07 | — | — | — | — | — | — | — | — | |
| cLSDMn=15002026.03 | 15.05 | — | — | — | — | — | — | — | — | |
| MicroBatchGANreported_in_original_paper=true2021.01 | 17.1 | — | — | — | — | — | — | — | — | |
| cLSDMn=5002026.03 | 17.72 | — | — | — | — | — | — | — | — | |
| cGANn=15002026.03 | 18.2 | — | — | — | — | — | — | — | — | |
| dLSDMn=15002026.03 | 18.43 | — | — | — | — | — | — | — | — | |
| cWGANn=15002026.03 | 18.44 | — | — | — | — | — | — | — | — | |
| LDMn=15002026.03 | 18.72 | — | — | — | — | — | — | — | — | |
| cLSDMn=250, Autoencoder architecture=WAE, Regularization/Loss=W12026.03 | 18.96 | — | — | — | — | — | — | — | — | |
| cLSDMn=2502026.03 | 19.5 | — | — | — | — | — | — | — | — | |
| DCGANreported_in_original_paper=false2021.01 | 19.86 | — | — | — | — | — | — | — | — | |
| QACGANEpochs=5, Seed=42, Total training time=7h 17m 36s2025.12 | 19.92 | — | — | — | 2.07 | — | — | 19.92 | 99 | |
| WGAN GPCategory=adversarial training2020.07 | 20.3 | — | — | — | — | — | — | — | — | |
| cLSDMn=1252026.03 | 20.49 | — | — | — | — | — | — | — | — | |
| dLSDMn=5002026.03 | 21.17 | — | — | — | — | — | — | — | — | |
| dLSDMn=250, Autoencoder architecture=VQVAE, Regularization/Loss=W12026.03 | 21.37 | — | — | — | — | — | — | — | — | |
| dLSDMn=2502026.03 | 21.41 | — | — | — | — | — | — | — | — | |
| GAN2021.06 | 21.5 | — | — | — | — | — | — | — | — | |
| cLSDMn=250, Autoencoder architecture=VQVAE, Regularization/Loss=W12026.03 | 21.59 | — | — | — | — | — | — | — | — | |
| dLSDMn=250, Autoencoder architecture=WAE, Regularization/Loss=W12026.03 | 22.17 | — | — | — | — | — | — | — | — | |
| D2GANreported_in_original_paper=false2021.01 | 22.2 | — | — | — | — | — | — | — | — | |
| cLSDMn=250, Autoencoder architecture=AE, Regularization/Loss=JS2026.03 | 22.37 | — | — | — | — | — | — | — | — | |
| LDMn=5002026.03 | 22.82 | — | — | — | — | — | — | — | — | |
| QACGANEpochs=10, Run index=2, Seed=2025, Total training time=14h 9m 1s2025.12 | 23.23 | — | — | — | 2.32 | — | — | 24.11 | 100 | |
| CWAECategory=AE based2020.07 | 23.6 | — | — | — | — | — | — | — | — | |
| SSAETime (seconds/epoch)=64.812023.06 | 23.65 | — | — | — | — | — | — | — | — | |
| LSGANreported_in_original_paper=false2021.01 | 23.8 | — | — | — | — | — | — | — | — | |
| dLSDMn=1252026.03 | 23.86 | — | — | — | — | — | — | — | — | |
| QACGANEpochs=10, Run index=1, Seed=42, Total training time=14h 2m 17s2025.12 | 23.91 | — | — | — | 2.29 | — | — | 35.96 | 99 | |
| ACGANEpochs=50, Total training time=0h 27m 42s2025.12 | 24.02 | — | — | — | 2.23 | — | — | — | 87.8 | |
| cVAEn=15002026.03 | 24.15 | — | — | — | — | — | — | — | — | |
| SAETime (seconds/epoch)=125.872023.06 | 24.72 | — | — | — | — | — | — | — | — | |
| LDMn=2502026.03 | 25.29 | — | — | — | — | — | — | — | — | |
| cWGANn=5002026.03 | 25.64 | — | — | — | — | — | — | — | — | |
| SWAECategory=AE based, Reference=Kolouri et al., 20182020.07 | 29.8 | — | — | — | — | — | — | — | — | |
| pLDMn=15002026.03 | 30.17 | — | — | — | — | — | — | — | — | |
| cGANn=5002026.03 | 30.19 | — | — | — | — | — | — | — | — | |
| dLSDMn=250, Autoencoder architecture=AE, Regularization/Loss=JS2026.03 | 30.25 | — | — | — | — | — | — | — | — | |
| LDMn=1252026.03 | 31.63 | — | — | — | — | — | — | — | — | |
| cVAEn=5002026.03 | 33.8 | — | — | — | — | — | — | — | — | |
| NDSM-SGM2025.12 | 36.1 | — | — | — | 8.93 | — | — | — | — | |
| VAE2021.06 | 39.84 | — | — | — | — | — | — | — | — | |
| VFA-GAN2021.06 | 41.24 | — | — | — | — | — | — | — | — | |
| cWGANn=2502026.03 | 41.25 | — | — | — | — | — | — | — | — | |
| PresGANreported_in_original_paper=true2021.01 | 42.02 | — | — | — | — | — | — | — | — | |
| ms-DRAECategory=AE based2020.07 | 43.6 | — | — | — | — | — | — | — | — | |
| cVAEn=2502026.03 | 47.13 | — | — | — | — | — | — | — | — | |
| PEepsilon=12026.01 | 48 | — | — | — | — | 27.9 | — | — | — | |
| Oursprivacy_budget_epsilon=10, privacy_budget_delta=1e-5, sigma_noise=0.1, gamma_recon=12026.01 | 49.41 | — | — | — | 9.74 | — | — | — | — | |
| cWGANn=1252026.03 | 52.38 | — | — | — | — | — | — | — | — | |
| DP-GAN-DPACprivacy_budget_epsilon=10, privacy_budget_delta=1e-52026.01 | 54.06 | — | — | — | 9.71 | — | — | — | — | |
| DPSinkhornprivacy_budget_epsilon=10, privacy_budget_delta=1e-52026.01 | 55.56 | — | — | — | — | — | — | — | — | |
| GS-WGANprivacy_budget_epsilon=10, privacy_budget_delta=1e-52026.01 | 61.34 | — | — | — | 9.23 | — | — | — | — | |
| cGANn=2502026.03 | 61.42 | — | — | — | — | — | — | — | — | |
| cVAEn=1252026.03 | 66.41 | — | — | — | — | — | — | — | — | |
| DRAGANreported_in_original_paper=false2021.01 | 66.96 | — | — | — | — | — | — | — | — | |
| cGANn=1252026.03 | 68.84 | — | — | — | — | — | — | — | — | |
| pLDMn=5002026.03 | 96.69 | — | — | — | — | — | — | — | — | |
| cLSDMn=250, Autoencoder architecture=AE, Regularization/Loss=KL2026.03 | 105.05 | — | — | — | — | — | — | — | — | |
| pLDMn=2502026.03 | 105.06 | — | — | — | — | — | — | — | — | |
| MAGTSampling Time (s)=0.252026.02 | 109.22 | — | — | — | — | — | — | — | — | |
| GANSampling Time (s)=0.192026.02 | 109.23 | — | — | — | — | — | — | — | — | |
| DDIMSampling Time (s)=320.632026.02 | 109.53 | — | — | — | — | — | — | — | — | |
| FMSampling Time (s)=30.182026.02 | 115 | — | — | — | — | — | — | — | — | |
| VFA2021.06 | 120.02 | — | — | — | — | — | — | — | — | |
| dLSDMn=250, Autoencoder architecture=AE, Regularization/Loss=KL2026.03 | 120.05 | — | — | — | — | — | — | — | — | |
| pLDMn=1252026.03 | 144.72 | — | — | — | — | — | — | — | — | |
| Flow-GAN (ADV)Category=adversarial training2020.07 | 155.6 | — | — | — | — | — | — | — | — | |
| SWFCategory=iterative2020.07 | 225.1 | — | — | — | — | — | — | — | — | |
| AttnFlow-iMapLevels=3, Steps=4, Channels=96, Parameters (MB)=46.032021.06 | — | — | — | 0.43 | — | — | — | — | — | |
| AttnFlow-iMapLevels=3, Steps=2, Channels=96, Parameters (MB)=23.782021.06 | — | — | — | 0.41 | — | — | — | — | — | |
| AttnFlow-iMapLevels=3, Steps=2, Channels=48, Parameters (MB)=8.942021.06 | — | — | — | 0.39 | — | — | — | — | — | |
| AttnFlow-iTransLevels=3, Steps=4, Channels=96, Parameters (MB)=46.252021.06 | — | — | — | 0.44 | — | — | — | — | — | |
| AttnFlow-iTransLevels=3, Steps=2, Channels=96, Parameters (MB)=23.892021.06 | — | — | — | 0.42 | — | — | — | — | — | |
| AttnFlow-iTransLevels=3, Steps=2, Channels=48, Parameters (MB)=9.052021.06 | — | — | — | 0.4 | — | — | — | — | — | |
| DCGAN2021.02 | — | — | — | — | 1.26 | — | — | — | — | |
| DO-DCGAN/CBackbone=DCGAN, Strategy=Continual Learning2021.02 | — | — | — | — | 1.43 | — | — | — | — | |
| DO-DCGAN/PBackbone=DCGAN, Strategy=Pruning2021.02 | — | — | — | — | 1.42 | — | — | — | — |