Image Generation on Binarized MNIST
Metric: nats (higher is better)
Metric: nats (higher is better)
| # | Model↕ | nats▼ | Extra Data | Paper | Date↕ | Code |
|---|---|---|---|---|---|---|
| 1 | NADE | 88.33 | No | A Deep and Tractable Density Estimator | 2013-10-07 | Code |
| 2 | MADE 2hl (32 orders) | 86.64 | No | MADE: Masked Autoencoder for Distribution Estima... | 2015-02-12 | Code |
| 3 | EoNADE 2hl (128 orders) | 85.1 | No | - | - | Code |
| 4 | EoNADE-5 2hl (128 orders) | 84.68 | No | Iterative Neural Autoregressive Distribution Est... | 2014-06-05 | Code |
| 5 | PixelCNN | 81.3 | No | Pixel Recurrent Neural Networks | 2016-01-25 | Code |
| 6 | PixelRNN | 79.2 | No | Pixel Recurrent Neural Networks | 2016-01-25 | Code |
| 7 | Efficient-VDVAE | 79.09 | No | Efficient-VDVAE: Less is more | 2022-03-25 | Code |
| 8 | BFN | 77.87 | No | Bayesian Flow Networks | 2023-08-14 | Code |
| 9 | Locally Masked PixelCNN (8 orders) | 77.58 | No | Locally Masked Convolution for Autoregressive Mo... | 2020-06-22 | Code |
| 10 | CR-NVAE | 76.93 | No | Consistency Regularization for Variational Auto-... | 2021-05-31 | Code |