Image Generation on ImageNet 64x64 resolution (test)
1.33FIDEDM RES101 GUIDED
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| EDM RES101 GUIDEDClassifier=Off-the-shelf ResNet-101, Sampling steps=256 SDE steps, Number of samples=500002023.10 | 1.33 | — | — | — | — | — | — | — | |
| EDM2-XLNFE=63, # param. (M)=11192024.10 | 1.33 | — | — | — | — | — | — | — | |
| RestartBase Sampler=EDM, Configuration=36, {(10, 3, 19.35, 40.79), (10, 3, 1.09, 1.92), (7, 6, 0.59, 1.09), (7, 6, 0.30, 0.59), (7, 25, 0.06, 0.30)}2023.06 | 1.36 | — | 623 | — | — | — | — | — | |
| RestartBase Sampler=EDM, Configuration=36, {(6, 1, 19.35, 40.79),(6, 1, 1.09, 1.92), (7, 6, 0.59, 1.09), (7, 6, 0.30, 0.59), (7, 25, 0.06, 0.30)}2023.06 | 1.39 | — | 535 | — | — | — | — | — | |
| EDM BASELINEClassifier=None, Sampling steps=256 SDE steps, Number of samples=500002023.10 | 1.41 | — | — | — | — | — | — | — | |
| RestartBase Sampler=EDM, Configuration=36, {(3, 1, 19.35, 40.79),(6, 1, 1.09, 1.92), (6, 5, 0.59, 1.09), (6, 5, 0.30, 0.59), (6, 20, 0.06, 0.30)}2023.06 | 1.41 | — | 385 | — | — | — | — | — | |
| RestartBase Sampler=EDM, Configuration=36, {(4, 1, 19.35, 40.79), (4, 1, 1.09, 1.92), (4, 5, 0.59, 1.09), (4, 5, 0.30, 0.59), (6, 6, 0.06, 0.30)}2023.06 | 1.46 | — | 203 | — | — | — | — | — | |
| RestartBase Sampler=EDM, Configuration=18, {(3, 1, 19.35, 40.79), (4, 1, 1.09, 1.92), (4, 5, 0.59, 1.09), (4, 5, 0.30, 0.59), (4, 10, 0.06, 0.30)}2023.06 | 1.51 | — | 165 | — | — | — | — | — | |
| SiDNFE=1, # param. (M)=2962024.10 | 1.52 | — | — | — | — | — | — | — | |
| EDM2-SNFE=63, # param. (M)=2802024.10 | 1.58 | — | — | — | — | — | — | — | |
| TCMNFE=2, # param. (M)=1119, Backbone=EDM2-XL2024.10 | 1.62 | — | — | — | — | — | — | — | |
| ECMNFE=2, # param. (M)=1119, Backbone=EDM2-XL2024.10 | 1.67 | — | — | — | — | — | — | — | |
| CAF + GAN (Ours)N=12024.11 | 1.69 | — | — | — | 62.03 | 64 | — | — | |
| RestartBase Sampler=EDM, Configuration=18, {(3, 1, 19.35, 40.79), (4, 1, 1.09, 1.92), (4, 4, 0.59, 1.09), (4, 1, 0.30, 0.59), (4, 4, 0.06, 0.30)}2023.06 | 1.71 | — | 99 | — | — | — | — | — | |
| CTMN=22024.11 | 1.73 | — | — | — | 64.29 | 57 | — | — | |
| MultiStep-CDNFE=2, # param. (M)=12002024.10 | 1.9 | — | — | — | — | — | — | — | |
| CTMN=12024.11 | 1.92 | — | — | — | 70.38 | 57 | — | — | |
| RestartBase Sampler=EDM, Configuration=18, {(5, 1, 19.35, 40.79),(5, 1, 1.09, 1.92), (5, 1, 0.59, 1.09), (5, 1, 0.06, 0.30)}2023.06 | 1.95 | — | 67 | — | — | — | — | — | |
| ADMN=2502024.11 | 2.07 | — | — | — | — | 63 | — | — | |
| StyleGAN-XLN=12024.11 | 2.09 | — | — | — | 82.35 | 52 | — | — | |
| TCMNFE=1, # param. (M)=1119, Backbone=EDM2-XL2024.10 | 2.2 | — | — | — | — | — | — | — | |
| EMD-16NFE=1, # param. (M)=2962024.10 | 2.2 | — | — | — | — | — | — | — | |
| TCMNFE=2, # param. (M)=280, Backbone=EDM2-S2024.10 | 2.31 | — | — | — | — | — | — | — | |
| RestartBase Sampler=EDM, Configuration=14, {(3, 1, 19.35, 40.79), (3, 1, 1.09, 1.92), (3, 1, 0.06, 0.30)}2023.06 | 2.38 | — | 39 | — | — | — | — | — | |
| EDMN=792024.11 | 2.44 | — | — | — | 48.88 | 67 | — | — | |
| ECMNFE=1, # param. (M)=1119, Backbone=EDM2-XL2024.10 | 2.49 | — | — | — | — | — | — | — | |
| DMD2 w/o GANNFE=1, # param. (M)=2962024.10 | 2.6 | — | — | — | — | — | — | — | |
| DMDN=12024.11 | 2.62 | — | — | — | — | — | — | — | |
| iCT-deepNFE=2, # param. (M)=5922024.10 | 2.77 | — | — | — | — | — | — | — | |
| ECMNFE=2, # param. (M)=280, Backbone=EDM2-S2024.10 | 2.79 | — | — | — | — | — | — | — | |
| TCMNFE=1, # param. (M)=280, Backbone=EDM2-S2024.10 | 2.88 | — | — | — | — | — | — | — | |
| iDDPMN=2502024.11 | 2.92 | — | — | — | — | 62 | — | — | |
| rKLPolicy=π(x, σ)2026.02 | 2.92 | — | 52.1 | — | — | — | 85.4 | 92.1 | |
| KLPolicy=π(x, σ)2026.02 | 2.96 | — | 51.8 | — | — | — | 85.4 | 92.3 | |
| Moment MatchingNFE=1, # param. (M)=4002024.10 | 3 | — | — | — | — | — | — | — | |
| EDMPolicy=None2026.02 | 3.01 | — | 350 | — | — | — | 85.8 | 91.8 | |
| DEISN=202024.11 | 3.1 | — | — | — | — | — | — | — | |
| MultiStep-CDNFE=1, # param. (M)=12002024.10 | 3.2 | — | — | — | — | — | — | — | |
| iCTNFE=2, # param. (M)=2962024.10 | 3.2 | — | — | — | — | — | — | — | |
| iCT-deepNFE=1, # param. (M)=5922024.10 | 3.25 | — | — | — | — | — | — | — | |
| DPM-solverN=202024.11 | 3.42 | — | — | — | — | — | — | — | |
| Wide-RN Robust JointInference Steps=2, Base Model=Consistency Distillation, Backbone=Wide-ResNet-50-22024.05 | 3.78 | — | — | — | 55.13 | — | 71 | 61 | |
| RN50 Robust JointInference Steps=2, Base Model=Consistency Distillation, Backbone=ResNet-502024.05 | 3.84 | — | — | — | 51.12 | — | 72 | 60 | |
| Moment MatchingNFE=2, # param. (M)=4002024.10 | 3.86 | — | — | — | — | — | — | — | |
| iCTNFE=1, # param. (M)=2962024.10 | 4.02 | — | — | — | — | — | — | — | |
| ECMNFE=1, # param. (M)=280, Backbone=EDM2-S2024.10 | 4.05 | — | — | — | — | — | — | — | |
| BigGAN-deepN=12024.11 | 4.06 | — | — | — | — | 48 | — | — | |
| RN50 Robust ClassifierInference Steps=2, Base Model=Consistency Distillation, Backbone=ResNet-502024.05 | 4.2 | — | — | — | 46.78 | — | 69 | 63 | |
| Wide-RN Non-Robust JointInference Steps=2, Base Model=Consistency Distillation, Backbone=Wide-ResNet-50-22024.05 | 4.66 | — | — | — | 46.37 | — | 68 | 63 | |
| Consistency DistillationInference Steps=22024.05 | 4.69 | — | — | — | 42.28 | — | 68 | 63 | |
| CDN=22024.11 | 4.7 | — | — | — | — | 64 | — | — | |
| Wide-RN Robust JointInference Steps=1, Base Model=Consistency Distillation, Backbone=Wide-ResNet-50-22024.05 | 4.84 | — | — | — | 58.73 | — | 70 | 60 | |
| RN50 Robust JointInference Steps=1, Base Model=Consistency Distillation, Backbone=ResNet-502024.05 | 4.95 | — | — | — | 51.98 | — | 72 | 58 | |
| TRACTNFE=2, # param. (M)=2962024.10 | 4.97 | — | — | — | — | — | — | — | |
| RN50 Robust ClassifierInference Steps=1, Base Model=Consistency Distillation, Backbone=ResNet-502024.05 | 5.55 | — | — | — | 44.65 | — | 68 | 61 | |
| Diff-InstructN=12024.11 | 5.57 | — | — | — | — | — | — | — | |
| Diff-InstructNFE=1, # param. (M)=2962024.10 | 5.57 | — | — | — | — | — | — | — | |
| CDN=12024.11 | 6.2 | — | — | — | 40.08 | 57 | — | — | |
| Consistency DistillationInference Steps=12024.05 | 6.2 | — | — | — | 39.87 | — | 67 | 63 | |
| Wide-RN Non-Robust JointInference Steps=1, Base Model=Consistency Distillation, Backbone=Wide-ResNet-50-22024.05 | 6.33 | — | — | — | 45.55 | — | 64 | 60 | |
| CAF (Ours)N=12024.11 | 6.52 | — | — | — | 37.45 | 62 | — | — | |
| DEISN=102024.11 | 6.65 | — | — | — | — | — | — | — | |
| TRACTN=12024.11 | 7.43 | — | — | — | — | — | — | — | |
| TRACTNFE=1, # param. (M)=2962024.10 | 7.43 | — | — | — | — | — | — | — | |
| DFNON=12024.11 | 7.83 | — | — | — | — | 61 | — | — | |
| DSNONFE=1, # param. (M)=3292024.10 | 7.83 | — | — | — | — | — | — | — | |
| DPM-solverN=102024.11 | 7.93 | — | — | — | — | — | — | — | |
| Wide-RN Robust JointInference Steps=2, Base Model=Consistency Training, Backbone=Wide-ResNet-50-22024.05 | 7.98 | — | — | — | 46.33 | — | 71 | 52 | |
| RN50 Robust JointInference Steps=2, Base Model=Consistency Training, Backbone=ResNet-502024.05 | 8.51 | — | — | — | 39.4 | — | 74 | 49 | |
| Wide-RN Robust JointInference Steps=1, Base Model=Consistency Training, Backbone=Wide-ResNet-50-22024.05 | 8.83 | — | — | — | 55.67 | — | 73 | 45 | |
| PDNFE=2, # param. (M)=2962024.10 | 8.95 | — | — | — | — | — | — | — | |
| RN50 Robust JointInference Steps=1, Base Model=Consistency Training, Backbone=ResNet-502024.05 | 9.6 | — | — | — | 50.8 | — | 76 | 39 | |
| RN50 Robust ClassifierInference Steps=2, Base Model=Consistency Training, Backbone=ResNet-502024.05 | 9.65 | — | — | — | 33.55 | — | 68 | 54 | |
| RN50 Robust ClassifierInference Steps=1, Base Model=Consistency Training, Backbone=ResNet-502024.05 | 10.71 | — | — | — | 36.38 | — | 70 | 47 | |
| DDPMN=2502024.11 | 11 | — | — | — | — | 58 | — | — | |
| CTN=22024.11 | 11.1 | — | — | — | — | 56 | — | — | |
| Consistency TrainingInference Steps=22024.05 | 11.12 | — | — | — | 27.28 | — | 68 | 54 | |
| Wide-RN Non-Robust JointInference Steps=2, Base Model=Consistency Training, Backbone=Wide-ResNet-50-22024.05 | 11.28 | — | — | — | 33.16 | — | 63 | 54 | |
| Wide-RN Non-Robust JointInference Steps=1, Base Model=Consistency Training, Backbone=Wide-ResNet-50-22024.05 | 12.3 | — | — | — | 35.62 | — | 64 | 46 | |
| CTN=12024.11 | 13 | — | — | — | — | 47 | — | — | |
| Consistency TrainingInference Steps=12024.05 | 13 | — | — | — | 28.83 | — | 71 | 41 | |
| DDIMN=502024.11 | 13.7 | — | — | — | — | 56 | — | — | |
| PDN=12024.11 | 15.39 | — | — | — | — | 62 | — | — | |
| PDNFE=1, # param. (M)=2962024.10 | 15.4 | — | — | — | — | — | — | — | |
| HSIVI-SMNFE=15, Backbone=(Nichol & Dhariwal, 2021)2023.10 | 15.49 | — | — | — | — | — | — | — | |
| LA-NPR-DDPMTimesteps=10002023.04 | 16.27 | 0.2 | — | — | — | — | — | — | |
| LA-SN-DDPMTimesteps=10002023.04 | 16.28 | 0.1 | — | — | — | — | — | — | |
| NPR-DDPMTimesteps=10002023.04 | 16.3 | — | — | — | — | — | — | — | |
| SN-DDPMTimesteps=10002023.04 | 16.3 | — | — | — | — | — | — | — | |
| SN-DDPMTimesteps=2002023.04 | 16.7 | — | — | — | — | — | — | — | |
| NPR-DDPMTimesteps=2002023.04 | 16.75 | — | — | — | — | — | — | — | |
| LA-SN-DDPMTimesteps=2002023.04 | 16.76 | -0.4 | — | — | — | — | — | — | |
| LA-NPR-DDPMTimesteps=2002023.04 | 16.83 | -0.5 | — | — | — | — | — | — | |
| SN-DDIMTimesteps=2002023.04 | 17.14 | — | — | — | — | — | — | — | |
| NPR-DDIMTimesteps=2002023.04 | 17.15 | — | — | — | — | — | — | — | |
| NPR-DDIMTimesteps=10002023.04 | 17.59 | — | — | — | — | — | — | — | |
| SN-DDIMTimesteps=10002023.04 | 17.6 | — | — | — | — | — | — | — | |
| SN-DDIMTimesteps=1002023.04 | 17.67 | — | — | — | — | — | — | — | |
| HSIVI-SMNFE=10, Backbone=(Nichol & Dhariwal, 2021)2023.10 | 17.67 | — | — | — | — | — | — | — | |
| NPR-DDIMTimesteps=1002023.04 | 17.71 | — | — | — | — | — | — | — |