Conditional Image Generation on ImageNet 256x256 1.0 (train val)
1.42FIDSiT-XL/2 + REPA
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| SiT-XL/2 + REPACFG=true2024.12 | 1.42 | 305.7 | 0.8 | 0.65 | — | — | — | |
| MAR-HTok. Model=KL, # Tokens=256, CFG=true2024.12 | 1.55 | 303.7 | 0.81 | 0.62 | 145.08 | 0.12 | 1.22 | |
| MDTv2-XL/2CFG=true2024.12 | 1.58 | 314.7 | 0.79 | 0.65 | 125.43 | 0.59 | — | |
| SiT-XL (SoftVQ-L)Tok. Model=SoftVQ-L, # Tokens=64, CFG=true2024.12 | 1.86 | 293.6 | 0.81 | 0.63 | 29.23 | 5.39 | 0.61 | |
| MAR-H (SoftVQ-BL)Tok. Model=SoftVQ-BL, # Tokens=64, CFG=true2024.12 | 1.93 | 289.4 | 0.8 | 0.61 | 86.55 | 0.89 | 0.65 | |
| TiTok-S-128Tok. Model=VQ, # Tokens=128, CFG=true2024.12 | 1.97 | 281.8 | — | — | 33.03 | 6.5 | 1.61 | |
| SiT-XL/2CFG=true2024.12 | 2.06 | 270.3 | 0.82 | 0.59 | 81.92 | 0.54 | — | |
| DiT-XL/2CFG=true2024.12 | 2.27 | 278.2 | 0.83 | 0.53 | 80.73 | 0.51 | — | |
| U-ViT-H/2Tok. Model=KL, # Tokens=1024, Sampler=DPM-Solver, Sampling Steps=50, CFG=true2024.12 | 2.29 | 263.9 | 0.82 | 0.57 | 128.89 | 0.98 | 0.62 | |
| MAR-HTok. Model=KL, # Tokens=256, CFG=false2024.12 | 2.35 | 227.8 | 0.79 | 0.62 | 145.08 | 0.12 | 1.22 | |
| SiT-XL (SoftVQ-L)Tok. Model=SoftVQ-L, # Tokens=32, CFG=true2024.12 | 2.44 | 310.6 | 0.77 | 0.63 | 14.73 | 10.11 | 0.74 | |
| MAR-H (SoftVQ-L)Tok. Model=SoftVQ-L, # Tokens=32, CFG=true2024.12 | 2.54 | 273.6 | 0.78 | 0.61 | 67.93 | 2.19 | 0.61 | |
| MAR-H (SoftVQ-BL)Tok. Model=SoftVQ-BL, # Tokens=64, CFG=false2024.12 | 2.81 | 218.3 | 0.78 | 0.62 | 86.55 | 0.89 | 0.65 | |
| DiT-XL (SoftVQ-L)Tok. Model=SoftVQ-L, # Tokens=64, CFG=true2024.12 | 2.93 | 268.5 | 0.84 | 0.55 | 28.81 | 4.51 | 0.61 | |
| LlamaGen-3BTok. Model=VQ, # Tokens=256, CFG=true2024.12 | 3.06 | 279.7 | 0.8 | 0.58 | 781.56 | 2.9 | 2.08 | |
| LDM-4Tok. Model=KL, # Tokens=4096, CFG=true2024.12 | 3.6 | 247.7 | 0.87 | 0.48 | 157.92 | 0.37 | 0.27 | |
| DiT-XL (SoftVQ-L)Tok. Model=SoftVQ-L, # Tokens=32, CFG=true2024.12 | 3.69 | 270.4 | 0.83 | 0.53 | 14.52 | 8.74 | 0.74 | |
| RQ-Trans.Tok. Model=RQ, # Tokens=256, CFG=false2024.12 | 3.8 | 323.7 | — | — | 908.91 | 7.85 | 3.2 | |
| MAR-H (SoftVQ-L)Tok. Model=SoftVQ-L, # Tokens=32, CFG=false2024.12 | 3.83 | 211.2 | 0.77 | 0.61 | 67.93 | 2.19 | 0.61 | |
| Taming-Trans.Tok. Model=VQ, # Tokens=256, CFG=false2024.12 | 5.2 | 290.3 | — | — | — | — | 7.94 | |
| MaskGITTok. Model=VQ, # Tokens=256, CFG=false2024.12 | 6.18 | 182.1 | 0.8 | 0.51 | — | — | 2.28 | |
| MAGETok. Model=VQ, # Tokens=256, CFG=false2024.12 | 6.93 | 195.8 | — | — | — | — | — | |
| LDM-4Tok. Model=KL, # Tokens=4096, CFG=false2024.12 | 10.56 | 103.5 | 0.71 | 0.62 | 157.92 | 0.37 | 0.27 |