Autoregressive Super-Resolution (64 to 256) on KF256 (test)
0.309RL2Heun
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| HeunNFE=10, category=Flow Matching teacher (multi-step)2026.05 | 0.309 | 67.8 | 3.29 | |
| RK5NFE=30, category=Flow Matching teacher (multi-step)2026.05 | 0.308 | 68.2 | 3.29 | |
| EulerNFE=5, category=Flow Matching teacher (multi-step)2026.05 | 0.304 | 68.9 | 3.26 | |
| DPM++NFE=30, category=Diffusion (multi-step)2026.05 | 0.293 | 68.1 | 4.93 | |
| DDIMNFE=30, category=Diffusion (multi-step)2026.05 | 0.291 | 68.8 | 4.75 | |
| sCM (distilled)NFE=1, category=Consistency Models (1-step)2026.05 | 0.287 | 71.7 | 1.34 | |
| sCM (trained)NFE=1, category=Consistency Models (1-step)2026.05 | 0.282 | 72.3 | 5.27 | |
| DM-FNOInput Resolution=64x64, Target Resolution=256x256, Size (NO + Decoder)=27.10M + 6.63M, Inference Time per Frame (s)=0.377 ± 0.0082026.04 | 0.27 | 72 | 6.7 | |
| DM-UNOInput Resolution=64x64, Target Resolution=256x256, Size (NO + Decoder)=26.83M + 6.63M, Inference Time per Frame (s)=0.376 ± 0.0082026.04 | 0.26 | 72 | 5.76 | |
| UNOInput Resolution=64x64, Target Resolution=256x2562026.04 | 0.22 | 71 | 1.76 | |
| FNOInput Resolution=64x64, Target Resolution=256x2562026.04 | 0.13 | 82 | 1 | |
| MENO-FNOInput Resolution=64x64, Target Resolution=256x256, Size (NO + Decoder)=27.10M + 6.63M, Inference Time per Frame (s)=0.030 ± 0.0042026.04 | 0.09 | 92 | 6.25 | |
| MENO-UNOInput Resolution=64x64, Target Resolution=256x256, Size (NO + Decoder)=26.83M + 6.63M, Inference Time per Frame (s)=0.030 ± 0.0042026.04 | 0.08 | 93 | 5.16 |