Generative Modeling on ImageNet 64x64 (downsampled)
1.41Bits Per DimensionSPN
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SPNquantization=5-bit2018.12 | 1.41 | — | — | |
| Glowquantization=5-bit2018.12 | 1.76 | — | — | |
| MaCow2023.01 | 3.69 | 2.91 | 8.05 | |
| MALIFlow Type=Continuous Flow (FFJORD)2021.02 | 3.71 | — | — | |
| Residual FlowFlow Type=Discrete Flow2021.02 | 3.76 | — | — | |
| GlowK=32, L=3, resolution=64x64, permutation=invertible 1 x 1 convolution2018.07 | 3.81 | — | — | |
| GlowFlow Type=Discrete Flow2021.02 | 3.81 | — | — | |
| Emerging2023.01 | 3.81 | 1.71 | 137.04 | |
| RNODEFlow Type=Continuous Flow (FFJORD)2021.02 | 3.83 | — | — | |
| CInC Flow2023.01 | 3.85 | 1.57 | 55.71 | |
| FInC Flow2023.01 | 3.88 | 1.43 | 2.11 | |
| RealNVPK=32, L=3, resolution=64x642018.07 | 3.98 | — | — | |
| RealNVPFlow Type=Discrete Flow2021.02 | 3.98 | — | — |