Conditional Image Generation on ImageNet 256x256 (train/val)
0.27Tok. rFIDLDM-4
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| LDM-4Gen. Model=Diffusion-based, Tok. Model=KL, # Tokens=40962024.12 | 0.27 | 157.92 | 0.37 | 10.56 | 103.5 | 3.6 | 247.7 | |
| MAR-HGen. Model=Auto-regressive, Tok. Model=SoftVQ-L, # Tokens=322024.12 | 0.61 | 67.93 | 2.08 | 3.83 | 211.2 | 2.54 | 273.6 | |
| DiT-XLGen. Model=Diffusion-based, Tok. Model=SoftVQ-L, # Tokens=642024.12 | 0.61 | 28.81 | 4.51 | 5.83 | 141.3 | 2.93 | 268.5 | |
| SiT-XLGen. Model=Diffusion-based, Tok. Model=SoftVQ-L, # Tokens=642024.12 | 0.61 | 29.23 | 5.39 | 5.35 | 151.2 | 1.86 | 293.6 | |
| DiT-XL/2Gen. Model=Diffusion-based, Tok. Model=KL, # Tokens=10242024.12 | 0.62 | 80.73 | 0.51 | 9.62 | 121.5 | 2.27 | 278.2 | |
| SiT-XL/2Gen. Model=Diffusion-based, Tok. Model=KL, # Tokens=10242024.12 | 0.62 | 81.92 | 0.54 | 8.3 | 131.7 | 2.06 | 270.3 | |
| SiT-XL/2 + REPAGen. Model=Diffusion-based, Tok. Model=KL, # Tokens=10242024.12 | 0.62 | 81.92 | 0.54 | 5.9 | 157.8 | 1.42 | 305.7 | |
| MAR-HGen. Model=Auto-regressive, Tok. Model=SoftVQ-BL, # Tokens=642024.12 | 0.65 | 86.55 | 0.89 | 2.81 | 218.3 | 1.93 | 289.4 | |
| DiT-XLGen. Model=Diffusion-based, Tok. Model=SoftVQ-BL, # Tokens=642024.12 | 0.65 | 28.81 | 4.59 | 6.53 | 131.9 | 3.11 | 268.3 | |
| SiT-XLGen. Model=Diffusion-based, Tok. Model=SoftVQ-BL, # Tokens=642024.12 | 0.65 | 29.23 | 5.42 | 5.8 | 143.5 | 1.88 | 287.9 | |
| DiT-XLGen. Model=Diffusion-based, Tok. Model=SoftVQ-BL, # Tokens=322024.12 | 0.68 | 14.52 | 8.81 | 9.22 | 115.8 | 3.78 | 266.7 | |
| SiT-XLGen. Model=Diffusion-based, Tok. Model=SoftVQ-BL, # Tokens=322024.12 | 0.68 | 14.73 | 10.12 | 7.67 | 133.9 | 2.44 | 308.8 | |
| DiT-XLGen. Model=Diffusion-based, Tok. Model=SoftVQ-L, # Tokens=322024.12 | 0.74 | 14.52 | 8.74 | 9.07 | 117.2 | 3.69 | 270.4 | |
| SiT-XLGen. Model=Diffusion-based, Tok. Model=SoftVQ-L, # Tokens=322024.12 | 0.74 | 14.73 | 10.11 | 7.59 | 137.4 | 2.44 | 310.6 | |
| DiT-XLGen. Model=Diffusion-based, Tok. Model=SoftVQ-B, # Tokens=642024.12 | 0.88 | 28.81 | 4.7 | 6.62 | 129.2 | 3.29 | 262.5 | |
| SiT-XLGen. Model=Diffusion-based, Tok. Model=SoftVQ-B, # Tokens=642024.12 | 0.88 | 29.23 | 5.51 | 5.98 | 138 | 1.78 | 279 | |
| DiT-XLGen. Model=Diffusion-based, Tok. Model=SoftVQ-B, # Tokens=322024.12 | 0.89 | 14.52 | 8.94 | 9.83 | 113.8 | 3.91 | 264.2 | |
| SiT-XLGen. Model=Diffusion-based, Tok. Model=SoftVQ-B, # Tokens=322024.12 | 0.89 | 14.73 | 10.28 | 7.99 | 129.3 | 2.51 | 301.3 | |
| MAR-HGen. Model=Auto-regressive, Tok. Model=KL, # Tokens=2562024.12 | 1.22 | 145.08 | 0.12 | 2.35 | 227.8 | 1.55 | 303.7 | |
| TiTok-S-128Gen. Model=Auto-regressive, Tok. Model=VQ, # Tokens=1282024.12 | 1.61 | 33.03 | 6.5 | — | — | 1.97 | 281.8 | |
| LlamaGen-3BGen. Model=Auto-regressive, Tok. Model=VQ, # Tokens=2562024.12 | 2.19 | 781.56 | 2.9 | — | — | 3.06 | 279.7 | |
| MaskGITGen. Model=Auto-regressive, Tok. Model=VQ, # Tokens=2562024.12 | 2.28 | — | — | 6.18 | 182.1 | — | — | |
| RQ-Trans.Gen. Model=Auto-regressive, Tok. Model=RQ, # Tokens=2562024.12 | 3.2 | 908.91 | 7.85 | 3.8 | 323.7 | — | — | |
| Taming-Trans.Gen. Model=Auto-regressive, Tok. Model=VQ, # Tokens=2562024.12 | 7.94 | — | — | 5.2 | 290.3 | — | — | |
| MAGEGen. Model=Auto-regressive, Tok. Model=VQ, # Tokens=2562024.12 | — | — | — | 6.93 | 195.8 | — | — | |
| MDTv2-XL/2Gen. Model=Diffusion-based2024.12 | — | 125.43 | 0.59 | 5.06 | 155.6 | 1.58 | 314.7 | |
| U-ViT-H/2Gen. Model=Diffusion-based, Sampling Steps=50, Sampling Method=DPM-Solver2024.12 | — | 128.89 | 0.98 | — | — | 2.29 | 263.9 |