# Image Classification on ImageNet (test) (Clean and Attack Accuracy Metrics)

Source: [Image Classification on ImageNet (test) (Clean and Attack Accuracy Metrics)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-test-clean-and-attack-accuracy-metrics>)

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

Data updated: 2026-02-27T08:00:05.717Z

Evaluation of Image Classification performance on the test split of ImageNet, reporting clean accuracy alongside various adversarial robustness metrics, including CAA scores and accuracy under semantic/full attacks.

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-clean-and-attack-accuracy-metrics.md?metric=Clean%20Accuracy>): Higher is better; unit: PERCENT.

- [CAA3a Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-test-clean-and-attack-accuracy-metrics.md?metric=CAA3a%20Accuracy>): Higher is better; unit: PERCENT.

- [CAA3b Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-test-clean-and-attack-accuracy-metrics.md?metric=CAA3b%20Accuracy>): Higher is better; unit: PERCENT.

- [CAA3c Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-test-clean-and-attack-accuracy-metrics.md?metric=CAA3c%20Accuracy>): Higher is better; unit: PERCENT.

- [Semantic Attacks (Rand.) Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-test-clean-and-attack-accuracy-metrics.md?metric=Semantic%20Attacks%20(Rand.)%20Accuracy>): Higher is better; unit: PERCENT.

- [Semantic Attacks (Sched.) Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-test-clean-and-attack-accuracy-metrics.md?metric=Semantic%20Attacks%20(Sched.)%20Accuracy>): Higher is better; unit: PERCENT.

- [Full Attacks (Rand.) Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-test-clean-and-attack-accuracy-metrics.md?metric=Full%20Attacks%20(Rand.)%20Accuracy>): Higher is better; unit: PERCENT.

- [Full Attacks (Sched.) Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-test-clean-and-attack-accuracy-metrics.md?metric=Full%20Attacks%20(Sched.)%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Clean Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| Normal | 76.1 | Not recorded | 68fa6ecc1e209131cea42343 |
| Madry | 62.4 | Not recorded | 68fa6ecc1e209131cea42343 |
| GAT-f | 60 | Not recorded | 68fa6ecc1e209131cea42343 |
| Fast-AT | 53.8 | Not recorded | 68fa6ecc1e209131cea42343 |

### Normal

Clean Accuracy: 76.1. Result ID: 69a14ef635dfcabb00e6277a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ecc1e209131cea42343.md>)

Recorded evaluation configuration: {"Training Setting":"†"}. This record does not assert independent reproduction.

### Madry

Clean Accuracy: 62.4. Result ID: 69a14ef635dfcabb00e6277b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ecc1e209131cea42343.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa665a1e209131cea0fbae.md>)

Recorded evaluation configuration: {"Training Setting":"†","Constraint":"l\_inf"}. This record does not assert independent reproduction.

### GAT-f

Clean Accuracy: 60. Result ID: 69a14ef635dfcabb00e6277d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ecc1e209131cea42343.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ecc1e209131cea42343.md>)

[Implementation](<https://hsiung.cc/CARBEN/>)

Recorded evaluation configuration: {"Training Setting":"†"}. This record does not assert independent reproduction.

### Fast-AT

Clean Accuracy: 53.8. Result ID: 69a14ef635dfcabb00e6277c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ecc1e209131cea42343.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66e51e209131cea12efc.md>)

Recorded evaluation configuration: {"Training Setting":"†","Constraint":"l\_inf"}. This record does not assert independent reproduction.

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