Class-conditional Image Generation on ImageNet 256
1.98FIDCNS
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| CNSModel=SiT-XL/2, CFG Scale=1.52026.05 | 1.98 | 257.68 | 0.81 | 0.6 | — | — | 4.46 | |
| CNSModel=JiT-H/16, CFG Scale=2.22026.05 | 2.03 | 65.89 | 0.79 | 0.62 | — | — | 6.98 | |
| SDEModel=SiT-XL/2, CFG Scale=1.52026.05 | 2.06 | 277.5 | 0.83 | 0.59 | — | — | 4.49 | |
| SDEModel=JiT-H/16, CFG Scale=2.22026.05 | 2.08 | 65.88 | 0.79 | 0.61 | — | — | 7.59 | |
| ODEModel=SiT-XL/2, CFG Scale=1.52026.05 | 2.15 | 258.09 | 0.81 | 0.6 | — | — | 4.6 | |
| ODEModel=JiT-H/16, CFG Scale=2.22026.05 | 3.92 | 62.86 | 0.76 | 0.59 | — | — | 7.72 | |
| CNSModel=JiT-B/16, CFG Scale=2.62026.05 | 4.19 | 60.27 | 0.81 | 0.47 | — | — | 9.01 | |
| CNSModel=JiT-B/16, CFG Scale=3.02026.05 | 4.39 | 63.22 | 0.83 | 0.45 | — | — | 9.48 | |
| SDEModel=JiT-B/16, CFG Scale=3.02026.05 | 4.54 | 63.02 | 0.82 | 0.45 | — | — | 11.3 | |
| ODEModel=JiT-B/16, CFG Scale=3.02026.05 | 5.83 | 61.47 | 0.81 | 0.43 | — | — | 8.71 | |
| Soft-Di[M]OSteps=12025.09 | 6.4 | 214.8 | 83 | 39 | 1.27 | 97 | — | |
| di4c-dSteps=42025.09 | 6.57 | 213.6 | — | — | — | — | — | |
| MaskGitSteps=162025.09 | 6.6 | 224.1 | 83 | 40 | 1.25 | 98 | — | |
| MaskGitSteps=82025.09 | 6.66 | 221.6 | 83 | 40 | 1.23 | 97 | — | |
| theta-trapezoidalSteps=642025.09 | 6.7 | — | — | — | — | — | — | |
| di4cSteps=42025.09 | 6.79 | 209.2 | — | — | — | — | — | |
| Di[M]OSteps=12025.09 | 6.91 | 214 | 83 | 38 | 1.26 | 97 | — | |
| theta-trapezoidalSteps=322025.09 | 7.1 | — | — | — | — | — | — | |
| ReDi1Steps=42025.09 | 7.58 | 228 | 87 | 46 | 1.33 | 98 | — | |
| HaltonSteps=82025.09 | 7.73 | 248.8 | 86 | 31 | 1.37 | 97 | — | |
| ReDi2Steps=42025.09 | 7.86 | 240 | 87 | 44 | 1.31 | 97 | — | |
| HaltonSteps=162025.09 | 8.73 | 283.2 | 89 | 28 | 1.46 | 98 | — | |
| SDTTSteps=42025.09 | 8.97 | 205 | 88 | 41 | 1.43 | 97 | — | |
| HaltonSteps=322025.09 | 9.62 | 296.2 | 90 | 28 | 1.47 | 98 | — | |
| MaskGitSteps=42025.09 | 10.73 | 192.3 | 75 | 31 | 1.01 | 92 | — | |
| ReDi3-distillSteps=12025.09 | 11.68 | 182 | 83 | 44 | 1.25 | 96 | — | |
| DEFARBackbone=SiT-B/4, Params=131M, Iters=500k, Hour=289, Sampling steps (NFE)=50, Classifier-Free Guidance (CFG)=true2026.06 | 28.34 | 58.94 | 0.551 | 0.524 | — | — | 7.84 | |
| DEFAR w/o ADRBackbone=SiT-B/4, Params=131M, Iters=500k, Hour=289, Sampling steps (NFE)=50, Classifier-Free Guidance (CFG)=true2026.06 | 28.71 | 60.43 | 0.557 | 0.532 | — | — | 8.25 | |
| DEFAR w/o FCBackbone=SiT-B/4, Params=131M, Iters=500k, Hour=228, Sampling steps (NFE)=50, Classifier-Free Guidance (CFG)=true2026.06 | 29.69 | 59.25 | 0.539 | 0.549 | — | — | 9 | |
| DEFARBackbone=SiT-B/4, Params=131M, Iters=250k, Hour=144, Sampling steps (NFE)=50, Classifier-Free Guidance (CFG)=true2026.06 | 29.81 | 58.81 | 0.531 | 0.532 | — | — | 7.35 | |
| IPBackbone=SiT-B/4, Params=131M, Iters=500k, Hour=166, Sampling steps (NFE)=50, Classifier-Free Guidance (CFG)=true2026.06 | 29.84 | 60.2 | 0.563 | 0.523 | — | — | 8.8 | |
| SDSSBackbone=SiT-B/4, Params=131M, Iters=500k, Hour=228, Sampling steps (NFE)=50, Classifier-Free Guidance (CFG)=true2026.06 | 30.29 | 58.69 | 0.549 | 0.534 | — | — | 8.95 | |
| SiT (origin)Backbone=SiT-B/4, Params=131M, Iters=500k, Hour=166, Sampling steps (NFE)=50, Classifier-Free Guidance (CFG)=true2026.06 | 31.64 | 55.65 | 0.542 | 0.524 | — | — | 8.44 | |
| MDSS (4steps)Backbone=SiT-B/4, Params=131M, Iters=500k, Hour=228, Sampling steps (NFE)=50, Classifier-Free Guidance (CFG)=true2026.06 | 31.9 | 56.89 | 0.544 | 0.527 | — | — | 10.05 | |
| DEFARBackbone=SiT-B/4, Params=131M, Iters=500k, Hour=289, Sampling steps (NFE)=50, Classifier-Free Guidance (CFG)=false2026.06 | 56.39 | 26.62 | 0.386 | 0.594 | — | — | 9.5 | |
| DEFAR w/o ADRBackbone=SiT-B/4, Params=131M, Iters=500k, Hour=289, Sampling steps (NFE)=50, Classifier-Free Guidance (CFG)=false2026.06 | 57.17 | 26.55 | 0.398 | 0.59 | — | — | 10.65 | |
| DEFAR w/o FCBackbone=SiT-B/4, Params=131M, Iters=500k, Hour=228, Sampling steps (NFE)=50, Classifier-Free Guidance (CFG)=false2026.06 | 59.77 | 25.81 | 0.377 | 0.598 | — | — | 12.77 | |
| DEFARBackbone=SiT-B/4, Params=131M, Iters=250k, Hour=144, Sampling steps (NFE)=50, Classifier-Free Guidance (CFG)=false2026.06 | 60.4 | 26.21 | 0.375 | 0.586 | — | — | 9.05 | |
| IPBackbone=SiT-B/4, Params=131M, Iters=500k, Hour=166, Sampling steps (NFE)=50, Classifier-Free Guidance (CFG)=false2026.06 | 60.45 | 26.05 | 0.404 | 0.578 | — | — | 12.49 | |
| SDSSBackbone=SiT-B/4, Params=131M, Iters=500k, Hour=228, Sampling steps (NFE)=50, Classifier-Free Guidance (CFG)=false2026.06 | 61.25 | 25.81 | 0.387 | 0.588 | — | — | 13.23 | |
| SiT (origin)Backbone=SiT-B/4, Params=131M, Iters=500k, Hour=166, Sampling steps (NFE)=50, Classifier-Free Guidance (CFG)=false2026.06 | 61.64 | 24.67 | 0.387 | 0.581 | — | — | 12.2 | |
| MDSS (4steps)Backbone=SiT-B/4, Params=131M, Iters=500k, Hour=228, Sampling steps (NFE)=50, Classifier-Free Guidance (CFG)=false2026.06 | 63.66 | 24.72 | 0.382 | 0.583 | — | — | 14.38 | |
| SDTTSteps=12025.09 | 90.4 | 14 | 31 | 13 | 0.21 | 34 | — | |
| MaskGitSteps=22025.09 | 91.35 | 13.4 | 18 | 16 | 0.09 | 12 | — |