Class-conditional Image Generation on ImageNet (train/val)
2.87FIDEDM-2-XXL-512
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| EDM-2-XXL-512Image Resolution=512, Conditioning=true2024.11 | 2.87 | — | 0.84 | 0.71 | 1.22 | 0.97 | 82.45 | 0.75 | |
| EDM-2-L-512Image Resolution=512, Conditioning=true2024.11 | 2.9 | — | 0.84 | 0.7 | 1.22 | 0.97 | 82.1 | 0.71 | |
| EDM-2-XL-512Image Resolution=512, Conditioning=true2024.11 | 2.92 | — | 0.84 | 0.71 | 1.21 | 0.97 | 83.23 | 0.77 | |
| MAR-H-256Image Resolution=256, Conditioning=true2024.11 | 3.11 | — | 0.82 | 0.72 | 1.07 | 0.96 | 81.37 | 0.64 | |
| MAR-L-256Image Resolution=256, Conditioning=true2024.11 | 3.3 | — | 0.82 | 0.71 | 1.1 | 0.96 | 81.8 | 0.56 | |
| EDM-2-M-512Image Resolution=512, Conditioning=true2024.11 | 3.3 | — | 0.85 | 0.69 | 1.24 | 0.97 | 82.99 | 0.65 | |
| EDM-2-S-512Image Resolution=512, Conditioning=true2024.11 | 3.33 | — | 0.85 | 0.67 | 1.26 | 0.97 | 80.25 | 0.59 | |
| LDM-4-GNparams=400M, Sampling steps=250 DDIM steps, Unconditional guidance scale=1.5, Training steps=178K, Batch size=12002021.12 | 3.6 | 247.67 | 0.87 | 0.48 | — | — | — | — | |
| MAR-B-256Image Resolution=256, Conditioning=true2024.11 | 3.79 | — | 0.83 | 0.67 | 1.18 | 0.96 | 83.03 | 0.45 | |
| EDM-2-XS-512Image Resolution=512, Conditioning=true2024.11 | 3.79 | — | 0.83 | 0.65 | 1.22 | 0.95 | 72.41 | 0.46 | |
| ADM-G, ADM-USampling steps=2 x 250 DDIM steps2021.12 | 3.85 | 221.72 | 0.84 | 0.53 | — | — | — | — | |
| LDM-4-GNparams=400M, Sampling steps=250 DDIM steps, Unconditional guidance scale=1.25, Training steps=178K, Batch size=12002021.12 | 3.95 | 178.22 | 0.81 | 0.55 | — | — | — | — | |
| ADM-GNparams=608M, Sampling steps=250 DDIM steps2021.12 | 4.59 | 186.7 | 0.82 | 0.52 | — | — | — | — | |
| CDMSampling steps=2 x 100 DDIM steps2021.12 | 4.88 | 158.71 | — | — | — | — | — | — | |
| VQGAN+TNparams=1.3B, Classifier rejection rate=0.052021.12 | 5.88 | 304.8 | — | — | — | — | — | — | |
| ADM-256Image Resolution=256, Conditioning=true2024.11 | 6.01 | — | 0.82 | 0.62 | 1.08 | 0.91 | 70.94 | 0.44 | |
| BigGan-deepNparams=340M2021.12 | 6.95 | 203.6 | 0.87 | 0.28 | — | — | — | — | |
| ImageBARTNparams=3.5B, Classifier rejection rate=0.052021.12 | 7.44 | — | — | — | — | — | — | — | |
| LDM-8-GNparams=506M, Sampling steps=200 DDIM steps, Classifier scale=10, Training steps=4.8M, Batch size=642021.12 | 7.76 | 209.52 | 0.84 | 0.35 | — | — | — | — | |
| LDM-8-GNparams=506M, Sampling steps=200 DDIM steps, Classifier scale=10, Training steps=2.9M, Batch size=642021.12 | 8.11 | 190.43 | 0.83 | 0.36 | — | — | — | — | |
| LDM-4Nparams=400M, Sampling steps=250 DDIM steps, Training steps=178K, Batch size=12002021.12 | 10.56 | 103.49 | 0.71 | 0.62 | — | — | — | — | |
| ADMNparams=554M, Sampling steps=250 DDIM steps2021.12 | 10.94 | 100.98 | 0.69 | 0.63 | — | — | — | — | |
| SR3Nparams=625M2021.12 | 11.3 | — | — | — | — | — | — | — | |
| LDM-8Nparams=395M, Sampling steps=200 DDIM steps, Training steps=4.8M, Batch size=642021.12 | 15.51 | 79.03 | 0.65 | 0.63 | — | — | — | — | |
| VQGAN+TNparams=1.3B2021.12 | 17.04 | 70.6 | — | — | — | — | — | — | |
| LDM-8Nparams=395M, Sampling steps=200 DDIM steps, Training steps=2.9M, Batch size=642021.12 | 17.41 | 72.92 | 0.65 | 0.62 | — | — | — | — | |
| ImageBARTNparams=3.5B2021.12 | 21.19 | — | — | — | — | — | — | — | |
| DiT-XL/2-256Image Resolution=256, Conditioning=true2024.11 | 22.15 | — | 0.94 | 0.34 | 1.58 | 0.85 | 126.96 | 0.23 | |
| DiT-XL/2-512Image Resolution=512, Conditioning=true2024.11 | 22.99 | — | 0.96 | 0.27 | 1.9 | 0.86 | 128.98 | 0.21 | |
| LDM-256Image Resolution=256, Conditioning=true2024.11 | 26.09 | — | 0.96 | 0.21 | 1.8 | 0.83 | 126.94 | 0.16 |