Generative Modeling on ImageNet 256x256
1.73FIDVAR-d30-re (Tian et al., 2024)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| VAR-d30-re (Tian et al., 2024)Backbone=Transformers, Params=2.0B, Steps=102025.10 | 1.73 | 350.2 | |
| DiT-XL/2-G (Peebles and Xie, 2023b)Backbone=Transformers, Params=675M, Steps=2502025.10 | 2.27 | 278.2 | |
| DAT (T = 110)Backbone=ConvNeXt-L-CvSt, Params=198M, Steps=362025.10 | 3.29 | 310.2 | |
| VAR-d16 (Tian et al., 2024)Backbone=Transformers, Params=310M, Steps=102025.10 | 3.3 | 274.4 | |
| LDM-4-G (Rombach et al., 2022)Backbone=U-Net, Params=400M, Steps=2502025.10 | 3.6 | 247.7 | |
| ADM-G (Dhariwal and Nichol, 2021)Backbone=U-Net, Params=608M, Steps=2502025.10 | 4.59 | 186.7 | |
| DAT (T = 65)Backbone=WRN-50-4, Params=223M, Steps=192025.10 | 4.94 | 358 | |
| DAT (T = 30)Backbone=ResNet-50, Params=26M, Steps=142025.10 | 5.28 | 319.3 | |
| EGC (Guo et al., 2023)Backbone=U-Net, Params=543M, Steps=10002025.10 | 6.05 | 231.3 | |
| DAT (T = 30)Backbone=WRN-50-4, Params=223M, Steps=172025.10 | 6.23 | 341 | |
| DAT (T = 15)Backbone=ResNet-50, Params=26M, Steps=142025.10 | 6.87 | 322.65 | |
| BigGAN-deep (Brock et al., 2018)Backbone=ResNet, Params=340M, Steps=12025.10 | 6.95 | 203.6 | |
| Standard AT (Salman et al., 2020)Backbone=ResNet-50, Params=26M, Steps=132025.10 | 15.12 | 286.2 | |
| Standard AT (Salman et al., 2020)Backbone=WRN-50-4, Params=223M, Steps=122025.10 | 37.33 | 260.2 | |
| Standard AT (Singh et al., 2023)Backbone=ConvNeXt-L-CvSt, Params=198M, Steps=02025.10 | 44.46 | 27.32 |