# Image Classification on ImageNet (test) (Adversarial Robustness L1/L2/Linf Metrics)

Source: [Image Classification on ImageNet (test) (Adversarial Robustness L1/L2/Linf Metrics)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-test-adversarial-robustness-l1-l2-linf-metrics>)

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

Data updated: 2026-02-26T05:39:16.973Z

Evaluation of Image Classification performance on the test split of ImageNet, reporting robustness metrics against adversarial attacks measured across L1, L2, and L-infinity norms, along with Union Accuracy.

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 (test) |
| Metric | Clean Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 4 of 4 |
| 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-test-adversarial-robustness-l1-l2-linf-metrics.md?metric=Clean%20Accuracy>): Higher is better; unit: PERCENT.

- [Robust Accuracy (L-inf)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-test-adversarial-robustness-l1-l2-linf-metrics.md?metric=Robust%20Accuracy%20(L-inf)>): Higher is better; unit: PERCENT.

- [Robust Accuracy (L2)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-test-adversarial-robustness-l1-l2-linf-metrics.md?metric=Robust%20Accuracy%20(L2)>): Higher is better; unit: PERCENT.

- [Robust Accuracy (L1)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-test-adversarial-robustness-l1-l2-linf-metrics.md?metric=Robust%20Accuracy%20(L1)>): Higher is better; unit: PERCENT.

- [Union Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-test-adversarial-robustness-l1-l2-linf-metrics.md?metric=Union%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Clean Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| E-AT | 68.4 | Not recorded | 68fa74e71e209131cea6497f |
| RAMP | 66 | Not recorded | 68fa74e71e209131cea6497f |
| E-AT | 58.2 | Not recorded | 68fa74e71e209131cea6497f |
| RAMP | 55.6 | Not recorded | 68fa74e71e209131cea6497f |

### E-AT

Clean Accuracy: 68.4. Result ID: 69937eaf3a6c3148491658d7.

[Reporting paper](<https://www.sota2.com/research/paper/68fa74e71e209131cea6497f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d351e209131cea386bb.md>)

Recorded evaluation configuration: {"Backbone":"XCIT-S-L","Uses extra data for pre-training":false,"Source Model":"Debenedetti and Troncoso-EPFL, 2022"}. This record does not assert independent reproduction.

### RAMP

Clean Accuracy: 66. Result ID: 69937eaf3a6c3148491658d8.

[Reporting paper](<https://www.sota2.com/research/paper/68fa74e71e209131cea6497f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74e71e209131cea6497f.md>)

[Implementation](<https://github.com/uiuc-focal-lab/RAMP>)

Recorded evaluation configuration: {"Backbone":"XCIT-S-L","Uses extra data for pre-training":false,"Source Model":"Debenedetti and Troncoso-EPFL, 2022"}. This record does not assert independent reproduction.

### E-AT

Clean Accuracy: 58.2. Result ID: 69937eaf3a6c3148491658d9.

[Reporting paper](<https://www.sota2.com/research/paper/68fa74e71e209131cea6497f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d351e209131cea386bb.md>)

Recorded evaluation configuration: {"Backbone":"RN-50-l\_inf","Uses extra data for pre-training":false,"Source Model":"Engstrom et al., 2019"}. This record does not assert independent reproduction.

### RAMP

Clean Accuracy: 55.6. Result ID: 69937eaf3a6c3148491658da.

[Reporting paper](<https://www.sota2.com/research/paper/68fa74e71e209131cea6497f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74e71e209131cea6497f.md>)

[Implementation](<https://github.com/uiuc-focal-lab/RAMP>)

Recorded evaluation configuration: {"Backbone":"RN-50-l\_inf","Uses extra data for pre-training":false,"Source Model":"Engstrom et al., 2019"}. This record does not assert independent reproduction.

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