Text-to-image Generation on MS-COCO (Efficiency and FID-30K)
4.92FID-30KAdaGen-MaskGIT-S
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AdaGen-MaskGIT-SParams=57M, Steps=162026.03 | 4.92 | 0.6 | |
| AdaGen-MaskGIT-SParams=57M, Steps=82026.03 | 5.29 | 0.3 | |
| U-ViTParams=58M, Steps=502026.03 | 5.48 | 2.2 | |
| MaskGIT-SParams=45M, Steps=162026.03 | 5.78 | 0.6 | |
| U-ViTParams=58M, Steps=162026.03 | 5.91 | 0.9 | |
| PartiTokenizer=Discrete, Params=20B, Training Data=4.8B, Infer Steps=2562025.03 | 7.23 | — | |
| U-NetParams=53M, Steps=502026.03 | 7.32 | — | |
| Make-A-SceneParams=4B, Steps=15362026.03 | 7.55 | — | |
| MuseTokenizer=Discrete, Params=3B, Training Data=460M, A100 Days=26882025.03 | 7.88 | — | |
| FridoParams=512M, Steps=2002026.03 | 8.97 | — | |
| DALL-E 2Tokenizer=Continuous, Params=4.2B, Training Data=650M, Infer Steps=2502025.03 | 10.39 | — | |
| LDMTokenizer=Continuous, Params=1.4B, A100 Days=6250, Infer Steps=2502025.03 | 12.64 | — | |
| VQ-DiffusionParams=370M, Steps=1002026.03 | 13.86 | 206.7 | |
| FARTokenizer=Continuous, Params=564M, Training Data=7.8M, A100 Days=24, Infer Steps=102025.03 | 13.91 | — | |
| CogView2Tokenizer=Discrete, Params=6B, Training Data=35M2025.03 | 17.5 | — | |
| DALL-ETokenizer=Discrete, Params=12B, Training Data=250M, Infer Steps=2562025.03 | 27.5 | — |