Generative Modeling on CIFAR-10 (test)
2.65NLL (bits/dim)VDM
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| VDMType=Diff2023.01 | 2.65 | — | — | |
| Sparse Transformer2021.01 | 2.8 | — | — | |
| δ-VAE2021.01 | 2.83 | — | — | |
| ScoreFlowhorizontal flipping=false2021.01 | 2.83 | — | — | |
| WS-VDVAEType=VAE, #Params=39m2023.01 | 2.83 | — | — | |
| PixelSNAIL2018.02 | 2.85 | — | — | |
| PixelSNAIL2021.01 | 2.85 | — | — | |
| Very Deep VAE2021.01 | 2.87 | — | — | |
| VDVAE*Type=VAE, #Params=39m2023.01 | 2.87 | — | — | |
| VDVAEType=VAE, #Params=39m2023.01 | 2.87 | — | — | |
| WS-VDVAEType=VAE, #Params=25m2023.01 | 2.88 | — | — | |
| Image Transformer 1D local (cat)bsize=2562018.02 | 2.9 | — | — | |
| Image Transformer 1D local (dmol)bsize=2562018.02 | 2.9 | — | — | |
| Image Transformer2021.01 | 2.9 | — | — | |
| NVAE2021.01 | 2.91 | — | — | |
| NVAEType=VAE, #Params=131m2023.01 | 2.91 | — | — | |
| Pixel CNN++2018.02 | 2.92 | — | — | |
| PIXELCNN++autoregressive components=true2019.02 | 2.92 | — | — | |
| PixelCNN++2021.01 | 2.92 | — | — | |
| VLAEautoregressive components=true2019.02 | 2.95 | — | — | |
| VFlow2021.01 | 2.98 | — | — | |
| Image Transformer 1D local (8l, cat)bsize=2562018.02 | 2.99 | — | — | |
| PixelRNN2016.06 | 3 | — | — | |
| Row Pixel RNN2018.02 | 3 | — | — | |
| PIXELRNNautoregressive components=true2019.02 | 3 | — | — | |
| Gated PixelCNN2016.06 | 3.03 | — | — | |
| Gated Pixel CNN2018.02 | 3.03 | — | — | |
| GATED PIXELCNNautoregressive components=true2019.02 | 3.03 | — | — | |
| Gated PixelCNN2021.01 | 3.03 | — | — | |
| FLOW++autoregressive components=false2019.02 | 3.08 | — | — | |
| BIVAautoregressive components=false, Number of layers (L)=15, Importance weighted samples (k)=10002019.02 | 3.08 | — | — | |
| Flow++2021.01 | 3.08 | — | — | |
| IAFVAEautoregressive components=true, Importance weighted samples (k)=10002019.02 | 3.12 | — | — | |
| BIVAautoregressive components=false, Number of layers (L)=15, Importance weighted samples (k)=12019.02 | 3.12 | — | — | |
| Image Transformer 1D local (8l, cat)bsize=642018.02 | 3.13 | — | — | |
| PixelCNN2016.06 | 3.14 | — | — | |
| Pixel CNN2018.02 | 3.14 | — | — | |
| IAFVAEautoregressive components=true, Importance weighted samples (k)=12019.02 | 3.15 | — | — | |
| BIVAautoregressive components=false, Number of layers (L)=10, Importance weighted samples (k)=12019.02 | 3.17 | — | — | |
| EBM-FCETraining objective=Flow Contrastive Estimation, Normalizing constant=Estimated2019.12 | 3.27 | — | — | |
| GLOWautoregressive components=false2019.02 | 3.35 | — | — | |
| Glow-MLETraining objective=Maximum Likelihood Estimation2019.12 | 3.35 | — | — | |
| DISCRETEVAE++autoregressive components=false2019.02 | 3.38 | — | — | |
| QC-NCSN++2022.09 | 3.38 | 34,900,000 | 0.0008 | |
| FFJORD2021.01 | 3.4 | — | — | |
| Glow-FCETraining objective=Flow Contrastive Estimation2019.12 | 3.45 | — | — | |
| U-NCSN++2022.09 | 3.46 | 188,000,000 | 0.0019 | |
| RIDE2016.06 | 3.47 | — | — | |
| Image Transformer 1D local (8l, cat)bsize=162018.02 | 3.47 | — | — | |
| REALNVPautoregressive components=false2019.02 | 3.49 | — | — | |
| Conv DRAW2016.06 | 3.58 | — | — | |
| CONVDRAWautoregressive components=true2019.02 | 3.58 | — | — | |
| Deep GMMs2016.06 | 4 | — | — | |
| DEEPGMMSautoregressive components=false2019.02 | 4 | — | — | |
| Image Transformer 1D local (8l, cat)bsize=82018.02 | 4.06 | — | — | |
| DRAW2016.06 | 4.13 | — | — | |
| Deep Diffusion2016.06 | 4.2 | — | — | |
| NICE2016.06 | 4.48 | — | — | |
| NICEautoregressive components=false2019.02 | 4.48 | — | — | |
| Multivariate Gaussian2016.06 | 4.7 | — | — | |
| C-NCSN++2022.09 | 5.91 | 0 | 0 | |
| Uniform Distribution2016.06 | 8 | — | — |