Class-conditional image synthesis on ImageNet 256x256
2.4FIDVDM++
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VDM++Params=2B, Steps=256·22024.01 | 2.4 | 225.3 | |
| simple diffusionParams=2B, It.×BS=1B, Steps=5122024.01 | 2.77 | 211.8 | |
| RINParams=410M, It.×BS=614M, Steps=10002024.01 | 4.51 | 161 | |
| MaskDiT/2Params=736M+VAE, It.×BS=2B, Steps=40·22024.01 | 5.69 | 178 | |
| MDT-XL/2Params=676M+VAE, It.×BS=1.7B, Steps=2502024.01 | 6.23 | 143 | |
| U-ViT-H/2Params=501M+VAE, It.×BS=512M, Steps=50·22024.01 | 6.58 | — | |
| HDiTParams=557M, It.×BS=742M, Steps=50·22024.01 | 6.92 | 135.2 | |
| DiT-XL/2Params=675M+VAE, It.×BS=1.8B, Steps=2502024.01 | 9.62 | 121.5 | |
| LDM-4Params=400M+VAE, It.×BS=214M, Steps=2502024.01 | 10.56 | 209.5 | |
| ADMParams=554M, It.×BS=507M, Steps=10002024.01 | 10.94 | 101 | |
| iDDPMSteps=2502024.01 | 32.5 | — |