Image Generation on ImageNet 256 x 256 (FID U/G)
6.99FID (U)3-Stream (+Semantic)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| 3-Stream (+Semantic)Params=466M, Epochs=4002026.04 | 6.99 | 3.11 | |
| LF-DiT-L (Semantic Only)Params=465M, Epochs=4002026.04 | 7.2 | 3.21 | |
| SiT-XL/2Params=675M, Epochs=14002026.04 | 8.3 | 2.62 | |
| DiT-XL/2Params=675M, Epochs=14002026.04 | 9.62 | 3.04 | |
| JiT-LParams=459M, Epochs=2002026.04 | 16.21 | 2.79 | |
| 2-Stream (Learnable Wavelet)Params=461M, Epochs=4002026.04 | 17.08 | 3.99 | |
| 2-Stream (Laplacian)Params=461M, Epochs=4002026.04 | 20.7 | 4.87 | |
| 2-Stream (Haar)Params=461M, Epochs=4002026.04 | 22.84 | 5.02 | |
| 1-Stream + REPAEpochs=2002026.04 | 36.01 | 16.03 | |
| 1-Stream (Pixel DiT Baseline)Epochs=2002026.04 | 37.03 | 16.09 |