# Image Classification on ImageNet-C (level 4)

Source: [Image Classification on ImageNet-C (level 4)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4>)

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

Data updated: 2026-02-26T06:06:53.137Z

Evaluation of Image Classification robustness on the ImageNet-C dataset, specifically focusing on corruption level 4 across various distortion types (e.g., brightness, fog, noise).

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 | Image Classification |
| Dataset (including any stated split) | ImageNet-C (level 4) |
| Metric | Top-1 Acc (Brightness) |
| Unit | PERCENT |
| Direction | Higher 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

- [Top-1 Acc (Brightness)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4.md?metric=Top-1%20Acc%20(Brightness)>): Higher is better; unit: PERCENT.

- [Top-1 Acc (Contrast)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4.md?metric=Top-1%20Acc%20(Contrast)>): Higher is better; unit: PERCENT.

- [Top-1 Acc (Defocus)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4.md?metric=Top-1%20Acc%20(Defocus)>): Higher is better; unit: PERCENT.

- [Top-1 Acc (Elastic)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4.md?metric=Top-1%20Acc%20(Elastic)>): Higher is better; unit: PERCENT.

- [Top-1 Acc (Fog)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4.md?metric=Top-1%20Acc%20(Fog)>): Higher is better; unit: PERCENT.

- [Top-1 Acc (Frost)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4.md?metric=Top-1%20Acc%20(Frost)>): Higher is better; unit: PERCENT.

- [Top-1 Acc (Gaussian Noise)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4.md?metric=Top-1%20Acc%20(Gaussian%20Noise)>): Higher is better; unit: PERCENT.

- [Top-1 Acc (Glass Blur)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4.md?metric=Top-1%20Acc%20(Glass%20Blur)>): Higher is better; unit: PERCENT.

- [Top-1 Acc (Impulse Noise)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4.md?metric=Top-1%20Acc%20(Impulse%20Noise)>): Higher is better; unit: PERCENT.

- [Top-1 Acc (JPEG Compression)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4.md?metric=Top-1%20Acc%20(JPEG%20Compression)>): Higher is better; unit: PERCENT.

- [Top-1 Acc (Motion Blur)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4.md?metric=Top-1%20Acc%20(Motion%20Blur)>): Higher is better; unit: PERCENT.

- [Top-1 Acc (Pixelate)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4.md?metric=Top-1%20Acc%20(Pixelate)>): Higher is better; unit: PERCENT.

- [Top-1 Acc (Shot Noise)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4.md?metric=Top-1%20Acc%20(Shot%20Noise)>): Higher is better; unit: PERCENT.

- [Top-1 Acc (Snow)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4.md?metric=Top-1%20Acc%20(Snow)>): Higher is better; unit: PERCENT.

- [Top-1 Acc (Zoom Blur)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4.md?metric=Top-1%20Acc%20(Zoom%20Blur)>): Higher is better; unit: PERCENT.

- [Avg Top-1 Acc (ImageNet-C L4)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-c-level-4.md?metric=Avg%20Top-1%20Acc%20(ImageNet-C%20L4)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Top-1 Acc (Brightness) | Result date | Report paper |
| --- | --- | --- | --- |
| APM | 79.2 | Not recorded | 68fa7b941e209131cea89411 |
| OpenCLIP VIT-L/14 | 74.2 | Not recorded | 68fa7b941e209131cea89411 |
| Baseline | 73.1 | Not recorded | 68fa7b941e209131cea89411 |
| Baseline | 73.1 | Not recorded | 68fa6d691e209131cea399fa |
| TTT-MAE | 72.7 | Not recorded | 68fa7b941e209131cea89411 |
| TTT-MAE | 72.7 | Not recorded | 68fa6d691e209131cea399fa |

### APM

Top-1 Acc (Brightness): 79.2. Result ID: 698b87e67a967eddb76f5c43.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7b941e209131cea89411.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b941e209131cea89411.md>)

[Implementation](<https://rajatmodi62.github.io/apm_project_page>)

Recorded evaluation configuration: {"P (leveraged pre-trained weights)":false}. This record does not assert independent reproduction.

### OpenCLIP VIT-L/14

Top-1 Acc (Brightness): 74.2. Result ID: 698b87e67a967eddb76f5c42.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7b941e209131cea89411.md>)

Recorded evaluation configuration: {"P (leveraged pre-trained weights)":false}. This record does not assert independent reproduction.

### Baseline

Top-1 Acc (Brightness): 73.1. Result ID: 698b87e67a967eddb76f5c40.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7b941e209131cea89411.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d691e209131cea399fa.md>)

[Implementation](<https://yossigandelsman.github.io/ttt_mae/index.html>)

Recorded evaluation configuration: {"P (leveraged pre-trained weights)":true}. This record does not assert independent reproduction.

### Baseline

Top-1 Acc (Brightness): 73.1. Result ID: 698f9fd73a6c314849fa9f62.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6d691e209131cea399fa.md>)

Recorded evaluation configuration: {"TTT steps":10}. This record does not assert independent reproduction.

### TTT-MAE

Top-1 Acc (Brightness): 72.7. Result ID: 698b87e67a967eddb76f5c41.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7b941e209131cea89411.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d691e209131cea399fa.md>)

[Implementation](<https://yossigandelsman.github.io/ttt_mae/index.html>)

Recorded evaluation configuration: {"P (leveraged pre-trained weights)":true}. This record does not assert independent reproduction.

### TTT-MAE

Top-1 Acc (Brightness): 72.7. Result ID: 698f9fd73a6c314849fa9f63.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6d691e209131cea399fa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d691e209131cea399fa.md>)

[Implementation](<https://yossigandelsman.github.io/ttt_mae/index.html>)

Recorded evaluation configuration: {"TTT steps":10}. This record does not assert independent reproduction.

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