# Image Classification on ImageNet (val) (PGD Robustness Accuracy)

Source: [Image Classification on ImageNet (val) (PGD Robustness Accuracy)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-val-pgd-robustness-accuracy>)

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

Data updated: 2026-02-27T06:22:47.019Z

Evaluation of Image Classification performance on the validation split of the ImageNet dataset, reporting accuracy under adversarial attack conditions using Projected Gradient Descent (PGD) with 10, 50, and 100 iterations.

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 (val) |
| Metric | Accuracy (Natural Images) |
| 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

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

- [PGD-10 Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-val-pgd-robustness-accuracy.md?metric=PGD-10%20Accuracy>): Higher is better; unit: PERCENT.

- [PGD-50 Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-val-pgd-robustness-accuracy.md?metric=PGD-50%20Accuracy>): Higher is better; unit: PERCENT.

- [PGD-100 Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-val-pgd-robustness-accuracy.md?metric=PGD-100%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy (Natural Images) | Result date | Report paper |
| --- | --- | --- | --- |
| Natural | 76.038 | Not recorded | 68fa66ad1e209131cea11a3f |
| Free Adversarial Training | 71.21 | Not recorded | 68fa66ad1e209131cea11a3f |
| Free Adversarial Training | 64.446 | Not recorded | 68fa66ad1e209131cea11a3f |
| Free Adversarial Training | 60.642 | Not recorded | 68fa66ad1e209131cea11a3f |
| Free Adversarial Training | 58.116 | Not recorded | 68fa66ad1e209131cea11a3f |

### Natural

Accuracy (Natural Images): 76.038. Result ID: 69a1381f35dfcabb00e5960f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66ad1e209131cea11a3f.md>)

Recorded evaluation configuration: {"Backbone":"ResNet-50","epsilon":"2"}. This record does not assert independent reproduction.

### Free Adversarial Training

Accuracy (Natural Images): 71.21. Result ID: 69a1381f35dfcabb00e59610.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66ad1e209131cea11a3f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66ad1e209131cea11a3f.md>)

[Implementation](<https://github.com/ashafahi/free_adv_train>)

Recorded evaluation configuration: {"m":"2","Backbone":"ResNet-50","epsilon":"2"}. This record does not assert independent reproduction.

### Free Adversarial Training

Accuracy (Natural Images): 64.446. Result ID: 69a1381f35dfcabb00e59611.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66ad1e209131cea11a3f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66ad1e209131cea11a3f.md>)

[Implementation](<https://github.com/ashafahi/free_adv_train>)

Recorded evaluation configuration: {"m":"4","Backbone":"ResNet-50","epsilon":"2"}. This record does not assert independent reproduction.

### Free Adversarial Training

Accuracy (Natural Images): 60.642. Result ID: 69a1381f35dfcabb00e59612.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66ad1e209131cea11a3f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66ad1e209131cea11a3f.md>)

[Implementation](<https://github.com/ashafahi/free_adv_train>)

Recorded evaluation configuration: {"m":"6","Backbone":"ResNet-50","epsilon":"2"}. This record does not assert independent reproduction.

### Free Adversarial Training

Accuracy (Natural Images): 58.116. Result ID: 69a1381f35dfcabb00e59613.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66ad1e209131cea11a3f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66ad1e209131cea11a3f.md>)

[Implementation](<https://github.com/ashafahi/free_adv_train>)

Recorded evaluation configuration: {"m":"8","Backbone":"ResNet-50","epsilon":"2"}. This record does not assert independent reproduction.

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