Image Generation on CelebA HQ 256
17.5FID (50k samples)CFM-DDIM
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CFM-DDIMSampling Steps=4-Step2026.05 | 17.5 | — | — | — | |
| Consistency TrainingSampling Steps=4-Step2026.05 | 19 | — | — | — | |
| CFM-DDIMSampling Steps=128-Step2026.05 | 19.2 | — | — | — | |
| DDIMSampling Steps=128-Step2026.05 | 23 | — | — | — | |
| CFM-DDIMSampling Steps=1-Step2026.05 | 24.9 | — | — | — | |
| Consistency TrainingSampling Steps=1-Step2026.05 | 33.2 | — | — | — | |
| Consistency DistillationSampling Steps=1-Step2026.05 | 38.2 | — | — | — | |
| Consistency DistillationSampling Steps=4-Step2026.05 | 39.6 | — | — | — | |
| Consistency TrainingSampling Steps=128-Step2026.05 | 53.7 | — | — | — | |
| Consistency DistillationSampling Steps=128-Step2026.05 | 59.5 | — | — | — | |
| DDIMSampling Steps=4-Step2026.05 | 123.4 | — | — | — | |
| DDIMSampling Steps=1-Step2026.05 | 132.2 | — | — | — | |
| ADMNFE=10002026.05 | — | — | — | 5.11 | |
| CCVFM-BNFE=51, L=50, Backbone=DiT-B (131M params), K=10000, r=128, classifier-free guidance=false2026.05 | — | — | — | 5.59 | |
| CCVFM-LNFE=51, L=50, Backbone=DiT-L (458M params), K=10000, r=128, classifier-free guidance=false2026.05 | — | — | — | 4.17 | |
| DDPMNFE=10002026.05 | — | — | — | 7.03 | |
| DiTParameters=32.99 M2026.03 | — | 21.08 | 0.0802 | — | |
| EDMNFE=792026.05 | — | — | — | 3.5 | |
| LDM-4NFE=2002026.05 | — | — | — | 5.11 | |
| LSGMNFE=1382026.05 | — | — | — | 7.22 | |
| PDE-SSM-DiTParameters=26.48 M2026.03 | — | 18.36 | 0.0537 | — |