Image Generation on CelebA-HQ (test)
1.6583FIDLatent DDPM
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Latent DDPMTrain time=13.15 h, Sample time=16.23 ms, Device=H1002026.05 | 1.6583 | 0.0017 | 48.2 | 1.5 | |
| MM-SOLDTrain time=0 h, Sample time=45.21 ms, Device=CPU2026.05 | 1.8025 | 0.002 | 57.6 | 0 | |
| σ-CFDMTrain time=0 h, Sample time=54.91 ms, Device=CPU2026.05 | 2.5255 | 0.0029 | 42 | 0.2 | |
| HiT-BResolution=256x2562021.06 | 3.39 | — | — | — | |
| StyleGANResolution=1024x1024, bCR=false2021.06 | 5.17 | — | — | — | |
| LDMResolution=256x256, Representation=Discrete2024.01 | 5.51 | — | — | — | |
| LSGMResolution=256x256, Representation=Discrete2024.01 | 7.22 | — | — | — | |
| DDMIResolution=256x256, Representation=Continuous, Domain Specificity=Domain-agnostic2024.01 | 7.25 | — | — | — | |
| RDUOT2023.11 | 7.86 | — | — | — | |
| PG-GANResolution=256x2562021.06 | 8.03 | — | — | — | |
| DDMIResolution=128x128, Representation=Continuous, Domain Specificity=Domain-agnostic2024.01 | 8.73 | — | — | — | |
| HiT-BResolution=1024x1024, bCR=false2021.06 | 8.83 | — | — | — | |
| UOTM2023.11 | 8.84 | — | — | — | |
| COCO-GANResolution=1024x1024, bCR=false2021.06 | 9.49 | — | — | — | |
| DDMIResolution=64x64, Representation=Continuous, Domain Specificity=Domain-agnostic2024.01 | 9.74 | — | — | — | |
| INR-GANResolution=256x256, Representation=Continuous, Domain Specificity=Domain-specific2024.01 | 10.3 | — | — | — | |
| DDMIResolution=384x384, Representation=Continuous, Domain Specificity=Domain-agnostic2024.01 | 10.44 | — | — | — | |
| VQGANResolution=256x2562021.06 | 10.7 | — | — | — | |
| CIPSResolution=256x256, Representation=Continuous, Domain Specificity=Domain-specific2024.01 | 11.4 | — | — | — | |
| MeanFlowBatch size=64, Training steps=400K, Backbone=DiT-B/22026.02 | 12.4 | — | — | — | |
| DPFResolution=64x64, Representation=Continuous, Domain Specificity=Domain-agnostic2024.01 | 13.2 | — | — | — | |
| GASPResolution=64x64, Representation=Continuous, Domain Specificity=Domain-agnostic2024.01 | 13.5 | — | — | — | |
| CIPSResolution=128x128, Representation=Continuous, Domain Specificity=Domain-specific2024.01 | 13.53 | — | — | — | |
| Long-Short Flow Map / DriftingKernel type=Laplacian, B parameter=1024, Batch size=1024, Training steps=100K, Backbone=DiT-B/22026.02 | 14.71 | — | — | — | |
| CIPSResolution=64x64, Representation=Continuous, Domain Specificity=Domain-specific2024.01 | 15.41 | — | — | — | |
| CIPSResolution=384x384, Representation=Continuous, Domain Specificity=Domain-specific2024.01 | 15.8 | — | — | — | |
| DC-VAEResolution=256x2562020.11 | 15.81 | — | — | — | |
| Long-Short Flow Map / DriftingKernel type=Gaussian, B parameter=256, Batch size=1024, Training steps=100K, Backbone=DiT-B/22026.02 | 16.56 | — | — | — | |
| BOSS-6steps=62023.12 | 18.67 | — | — | — | |
| Long-Short Flow Map / DriftingKernel type=Laplacian, B parameter=256, Batch size=256, Training steps=100K, Backbone=DiT-B/22026.02 | 18.87 | — | — | — | |
| GASPResolution=128x128, Representation=Continuous, Domain Specificity=Domain-agnostic2024.01 | 19.2 | — | — | — | |
| StyleALAEResolution=256x2562020.11 | 19.21 | — | — | — | |
| StyleALAEResolution=256x2562021.06 | 19.21 | — | — | — | |
| Long-Short Flow Map / DriftingKernel type=Laplacian, B parameter=64, learning rate=mixed, Batch size=64, Training steps=400K, Backbone=DiT-B/22026.02 | 19.36 | — | — | — | |
| VAEBMResolution=256x2562021.06 | 20.38 | — | — | — | |
| Shortcut ModelsBatch size=64, Training steps=400K, Backbone=DiT-B/22026.02 | 20.5 | — | — | — | |
| NCP-VAEResolution=256x2562020.11 | 24.69 | — | — | — | |
| Long-Short Flow Map / DriftingKernel type=Gaussian, B parameter=64, Batch size=256, Training steps=100K, Backbone=DiT-B/22026.02 | 24.7 | — | — | — | |
| Long-Short Flow Map / DriftingKernel type=Laplacian, B parameter=64, Batch size=64, Training steps=100K, Backbone=DiT-B/22026.02 | 28.12 | — | — | — | |
| GEMResolution=64x64, Representation=Continuous, Domain Specificity=Domain-agnostic2024.01 | 30.4 | — | — | — | |
| Consistency ModelsBatch size=64, Training steps=400K, Backbone=DiT-B/22026.02 | 33.2 | — | — | — | |
| Distill-6-Reflowsteps=62023.12 | 34.56 | — | — | — | |
| NVAEResolution=256x2562020.11 | 40.26 | — | — | — | |
| FunctaResolution=64x64, Representation=Continuous, Domain Specificity=Domain-agnostic2024.01 | 40.4 | — | — | — | |
| Uniform-Reflowtype=reflow2023.12 | 43.57 | — | — | — |