Image Generation on CelebA-HQ 256x256 (test)
5.11FIDLDM-4
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LDM-4Type=Latent diffusion2025.11 | 5.11 | — | |
| Teacher2024.09 | 5.69 | 10.02 | |
| UNCSN++ (RVE) + STSoft Truncation=true, Backbone=UNCSN++, Variant=RVE2021.06 | 7.16 | — | |
| LSGM2022.01 | 7.22 | — | |
| LSGMType=Latent score2025.11 | 7.22 | — | |
| NCSN++ cont.Variant=deep, VE2021.06 | 7.23 | — | |
| SDEType=Score-based2025.11 | 7.23 | — | |
| PG-GANParams=46.1 M, FLOPs=14.1 G2021.12 | 8 | — | |
| DKDMmode=Dynamic Iterative Distillation2024.09 | 8.69 | 12.5 | |
| Data-Based Training2024.09 | 9.09 | 12.1 | |
| VQGAN + Transformer2022.01 | 10.2 | — | |
| VQ-GAN + TransformerParams=802 M, FLOPs=102 G2021.12 | 10.2 | — | |
| GAFType=Endpoint regression2025.11 | 10.32 | — | |
| DDIMT=100, Params=114 M, FLOPs=124 G2021.12 | 10.9 | — | |
| DiffuseVAET (Diffusion steps)=1000, GMM=100, FID evaluation sample size=10k2022.01 | 11.28 | — | |
| VQ-DDMK=1024, ReFiT=true, Params=117 M, FLOPs=1.06 G2021.12 | 13.2 | — | |
| Data-Limited TrainingData percentage=20%2024.09 | 14.49 | 17.08 | |
| Data-Limited TrainingData percentage=15%2024.09 | 14.89 | 16.98 | |
| Data-Limited TrainingData percentage=5%2024.09 | 15.07 | 17.64 | |
| Data-Limited TrainingData percentage=10%2024.09 | 15.23 | 17.53 | |
| Data-Free TrainingData percentage=0%2024.09 | 15.36 | 17.56 | |
| DC-VAE2021.12 | 15.8 | — | |
| DCVAE2022.01 | 15.81 | — | |
| D2C2022.01 | 18.74 | — | |
| VQ-DDMK=512, ReFiT=true, Params=117 M, FLOPs=1.04 G2021.12 | 18.8 | — | |
| VAEBM2022.01 | 20.38 | — | |
| VAEBMParams=127 M, FLOPs=8.22 G2021.12 | 20.4 | — | |
| VQ-DDMK=1024, ReFiT=false, Params=117 M, FLOPs=1.06 G2021.12 | 22.6 | — | |
| NCP-VAE2022.01 | 24.8 | — | |
| NVAE2022.01 | 40.26 | — | |
| NVAEParams=1.26 G, FLOPs=185 G2021.12 | 40.3 | — | |
| GLOWParams=220 M, FLOPs=540 G2021.12 | 60.9 | — | |
| GlowType=Flow2025.11 | 68.93 | — | |
| VAE BaselineGMM=100, FID evaluation sample size=10k2022.01 | 97.07 | — |