Class-conditional Image Generation on ImageNet (train)
3.6FIDLDM-CFG
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LDM-CFG#Param=400.9M, MACs=99.8G2022.11 | 3.6 | — | — | |
| ADM-G#Param=553.8+54.1M, MACs=1114.2+72.2G2022.11 | 4.6 | — | — | |
| LDM#Param=400.9M, MACs=99.8G2022.11 | 10.6 | — | — | |
| Spectral Diffusion (Ours-CFG)#Param=45.4M (8.8x), MACs=9.9G (10.1x)2022.11 | 10.6 | — | — | |
| ADM#Param=553.8M, MACs=1114.2G2022.11 | 10.9 | — | — | |
| IDDPM#Param=273.1M, MACs=1416.3G2022.11 | 12.3 | — | — | |
| Lite-LDM-CFG#Param=47.0M (8.5x), MACs=11.1G (9.0x)2022.11 | 20.1 | — | — | |
| JiTEpochs=802026.02 | — | 25.18 | 5.64 | |
| JiT+REPAEpochs=802026.02 | — | 18.6 | 4.57 | |
| LF-DiT D2V2Epochs=80, Backbone=D2V22026.02 | — | 12.46 | 5.45 | |
| LF-DiT DINOv2Epochs=80, Backbone=DINOv22026.02 | — | 9.76 | 4.18 |