Conditional Image Generation on ImageNet-1K (train/val)
3.25FIDMergeAR
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MergeARType=AR, Tokenizer=MergeVQ, Parameters=343M2026.05 | 3.25 | 253.8 | |
| LEASEType=Mask., Tokenizer=VQGAN, Parameters=117M+79M2026.05 | 3.72 | 179.09 | |
| DiGITType=AR, Tokenizer=VQGAN, Parameters=219M2026.05 | 4.79 | 142.87 | |
| MaskGITType=Mask., Tokenizer=VQGAN, Parameters=177M2026.05 | 6.18 | 182.1 | |
| MAGEType=Mask., Tokenizer=VQGAN, Parameters=117M+113M2026.05 | 6.93 | 195.8 | |
| REPA (SiT-L/2)Type=Diff., Parameters=458M2026.05 | 9.9 | 111.9 |