# Density Estimation on ImageNet 32x32 downsampled (val)

Source: [Density Estimation on ImageNet 32x32 downsampled (val)](<https://www.sota2.com/research/sota/density-estimation-on-imagenet-32x32-downsampled-val>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-02-26T05:53:17.316Z

Evaluation of Density Estimation performance on the validation split of the ImageNet dataset downsampled to 32x32 resolution, typically measured using Negative Log-Likelihood (bits/dim).

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Density Estimation |
| Dataset (including any stated split) | ImageNet 32x32 downsampled (val) |
| Metric | NLL (bits/dim) |
| Unit | SCALAR |
| Direction | Lower is better |
| Results loaded | 6 of 6 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [NLL (bits/dim)](<https://www.sota2.com/research/sota/density-estimation-on-imagenet-32x32-downsampled-val.md?metric=NLL%20(bits%2Fdim)>): Lower is better; unit: SCALAR.

- [KL Divergence (bits/dim)](<https://www.sota2.com/research/sota/density-estimation-on-imagenet-32x32-downsampled-val.md?metric=KL%20Divergence%20(bits%2Fdim)>): Lower is better; unit: SCALAR.

## Recorded results

| Method | NLL (bits/dim) | Result date | Report paper |
| --- | --- | --- | --- |
| ImageTransformer | 3.77 | Not recorded | 68fa654e1e209131cea09842 |
| Our Decoder baseline | 3.77 | Not recorded | 68fa654e1e209131cea09842 |
| PixelSNAIL | 3.8 | Not recorded | 68fa654e1e209131cea09842 |
| Gated PixelCNN | 3.83 | Not recorded | 68fa654e1e209131cea09842 |

### ImageTransformer

NLL (bits/dim): 3.77. Result ID: 698ca6093ce6cd2163e112a4.

[Reporting paper](<https://www.sota2.com/research/paper/68fa654e1e209131cea09842.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa64831e209131cea050c0.md>)

Recorded evaluation configuration: {"Model Category":"Autoregressive Models"}. This record does not assert independent reproduction.

### Our Decoder baseline

NLL (bits/dim): 3.77. Result ID: 698ca6093ce6cd2163e112a6.

[Reporting paper](<https://www.sota2.com/research/paper/68fa654e1e209131cea09842.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa654e1e209131cea09842.md>)

Recorded evaluation configuration: {"Model Category":"Autoregressive Models"}. This record does not assert independent reproduction.

### PixelSNAIL

NLL (bits/dim): 3.8. Result ID: 698ca6093ce6cd2163e112a5.

[Reporting paper](<https://www.sota2.com/research/paper/68fa654e1e209131cea09842.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63e81e209131cea015ac.md>)

Recorded evaluation configuration: {"Model Category":"Autoregressive Models"}. This record does not assert independent reproduction.

### Gated PixelCNN

NLL (bits/dim): 3.83. Result ID: 698ca6093ce6cd2163e112a3.

[Reporting paper](<https://www.sota2.com/research/paper/68fa654e1e209131cea09842.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa62b81e209131ce9fad7e.md>)

Recorded evaluation configuration: {"Model Category":"Autoregressive Models"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
