# Image Classification on ImageNet (Clean and PGD Robustness Evaluation)

Source: [Image Classification on ImageNet (Clean and PGD Robustness Evaluation)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-clean-and-pgd-robustness-evaluation>)

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

Data updated: 2026-02-26T05:39:17.562Z

Evaluation of Image Classification performance on the standard ImageNet dataset, focusing on clean accuracy alongside various Projected Gradient Descent (PGD) adversarial robustness metrics and associated iteration/time reduction statistics.

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 |
| Metric | Clean Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 5 of 5 |
| 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

- [Clean Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-clean-and-pgd-robustness-evaluation.md?metric=Clean%20Accuracy>): Higher is better; unit: PERCENT.

- [PGD Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-clean-and-pgd-robustness-evaluation.md?metric=PGD%20Accuracy>): Higher is better; unit: PERCENT.

- [PGDCD Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-clean-and-pgd-robustness-evaluation.md?metric=PGDCD%20Accuracy>): Higher is better; unit: PERCENT.

- [PGD Iterations (PGD)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-clean-and-pgd-robustness-evaluation.md?metric=PGD%20Iterations%20(PGD)>): Higher is better; unit: SCALAR.

- [PGD Iterations (PGDCD)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-clean-and-pgd-robustness-evaluation.md?metric=PGD%20Iterations%20(PGDCD)>): Higher is better; unit: SCALAR.

- [PGD Iterations Reduction](<https://www.sota2.com/research/sota/image-classification-on-imagenet-clean-and-pgd-robustness-evaluation.md?metric=PGD%20Iterations%20Reduction>): Higher is better; unit: PERCENT.

- [Time Reduction](<https://www.sota2.com/research/sota/image-classification-on-imagenet-clean-and-pgd-robustness-evaluation.md?metric=Time%20Reduction>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Clean Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| MixedNUTS | 81.11 | Not recorded | 68fa779c1e209131cea7456a |
| Swin-L | 78.19 | Not recorded | 68fa779c1e209131cea7456a |
| ConvNeXt-L | 77.48 | Not recorded | 68fa779c1e209131cea7456a |
| ViT-B-ConvStem | 76.12 | Not recorded | 68fa779c1e209131cea7456a |
| XCiT-S12 | 72.54 | Not recorded | 68fa779c1e209131cea7456a |

### MixedNUTS

Clean Accuracy: 81.11. Result ID: 699550543badfaffe6400215.

[Reporting paper](<https://www.sota2.com/research/paper/68fa779c1e209131cea7456a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa753c1e209131cea66a1e.md>)

Recorded evaluation configuration: {"Model":"(e)"}. This record does not assert independent reproduction.

### Swin-L

Clean Accuracy: 78.19. Result ID: 699550543badfaffe6400212.

[Reporting paper](<https://www.sota2.com/research/paper/68fa779c1e209131cea7456a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6eb01e209131cea41754.md>)

Recorded evaluation configuration: {"Model":"(b)"}. This record does not assert independent reproduction.

### ConvNeXt-L

Clean Accuracy: 77.48. Result ID: 699550543badfaffe6400211.

[Reporting paper](<https://www.sota2.com/research/paper/68fa779c1e209131cea7456a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6eb01e209131cea41754.md>)

Recorded evaluation configuration: {"Model":"(a)"}. This record does not assert independent reproduction.

### ViT-B-ConvStem

Clean Accuracy: 76.12. Result ID: 699550543badfaffe6400214.

[Reporting paper](<https://www.sota2.com/research/paper/68fa779c1e209131cea7456a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71011e209131cea4e7c0.md>)

[Implementation](<https://github.com/nmndeep/revisiting-at>)

Recorded evaluation configuration: {"Model":"(d)"}. This record does not assert independent reproduction.

### XCiT-S12

Clean Accuracy: 72.54. Result ID: 699550543badfaffe6400213.

[Reporting paper](<https://www.sota2.com/research/paper/68fa779c1e209131cea7456a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e741e209131cea3ff08.md>)

Recorded evaluation configuration: {"Model":"(c)"}. This record does not assert independent reproduction.

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