Generative Modeling on ImageNet 32x32 (test)
3.76NLLScoreFlow
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| ScoreFlow2021.01 | 3.76 | — | — | — | |
| Image Transformer2021.01 | 3.77 | — | — | — | |
| δ-VAE2021.01 | 3.77 | — | — | — | |
| Very Deep VAE2021.01 | 3.8 | — | — | — | |
| PixelSNAIL2021.01 | 3.8 | — | — | — | |
| Gated PixelCNN2021.01 | 3.83 | — | — | — | |
| VFlow2021.01 | 3.83 | — | — | — | |
| QC-NCSN++2022.09 | 3.83 | — | 11,300,000 | 0.0005 | |
| Flow++2021.01 | 3.86 | — | — | — | |
| NVAE2021.01 | 3.92 | — | — | — | |
| U-NCSN++2022.09 | 3.96 | — | 20,500,000 | 0.0007 | |
| C-NCSN++2022.09 | 5.1 | — | 0 | 0 | |
| NVAEType=VAE, #Params=268m2023.01 | — | 3.92 | — | — | |
| VDMType=Diff2023.01 | — | 3.72 | — | — | |
| VDVAEType=VAE, #Params=119m2023.01 | — | 3.8 | — | — | |
| VDVAE*Type=VAE, #Params=119m2023.01 | — | 3.67 | — | — | |
| WS-VDVAEType=VAE, #Params=55m2023.01 | — | 3.68 | — | — | |
| WS-VDVAEType=VAE, #Params=85m2023.01 | — | 3.65 | — | — |