# Image Classification on ImageNet (val) (Specialized Adversarial Robustness)

Source: [Image Classification on ImageNet (val) (Specialized Adversarial Robustness)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-val-specialized-adversarial-robustness>)

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

Data updated: 2026-02-27T11:16:50.087Z

Evaluation of Image Classification performance on the validation split of the ImageNet dataset, focusing on specialized adversarial robustness metrics including Cross-Modal PGD/AA, Targeted PGD/AA, and Label-free PGD/AA evaluations.

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 | Classification |
| Dataset (including any stated split) | ImageNet (val) |
| 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-val-specialized-adversarial-robustness.md?metric=Clean%20Accuracy>): Higher is better; unit: PERCENT.

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

- [Cross-Modal AA Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-val-specialized-adversarial-robustness.md?metric=Cross-Modal%20AA%20Accuracy>): Higher is better; unit: PERCENT.

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

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

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

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

- [Label-free AA Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-val-specialized-adversarial-robustness.md?metric=Label-free%20AA%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Clean Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| CLIP | 63.9 | Not recorded | 69799806a2b28e761d87cb52 |
| CSR | 62.5 | Not recorded | 69799806a2b28e761d87cb52 |
| FARE | 48 | Not recorded | 69799806a2b28e761d87cb52 |
| TTC | 40.9 | Not recorded | 69799806a2b28e761d87cb52 |

### CLIP

Clean Accuracy: 63.9. Result ID: 69a17d1735dfcabb00e760e3.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69581e209131cea21ad0.md>)

[Implementation](<https://github.com/OpenAI/CLIP>)

Recorded evaluation configuration: {"Perturbation budget":"4/255","Evaluation protocol":"Zero-shot"}. This record does not assert independent reproduction.

### CSR

Clean Accuracy: 62.5. Result ID: 69a17d1735dfcabb00e760e6.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

Recorded evaluation configuration: {"Perturbation budget":"4/255","Protocol":"Test-Time Defense"}. This record does not assert independent reproduction.

### FARE

Clean Accuracy: 48. Result ID: 69a17d1735dfcabb00e760e4.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738c1e209131cea5d24a.md>)

Recorded evaluation configuration: {"Perturbation budget":"4/255","Protocol":"Adversarial Fine-tuning"}. This record does not assert independent reproduction.

### TTC

Clean Accuracy: 40.9. Result ID: 69a17d1735dfcabb00e760e5.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa774a1e209131cea724cc.md>)

[Implementation](<https://www.sota2.com/here>)

Recorded evaluation configuration: {"Perturbation budget":"4/255","Protocol":"Test-Time Defense"}. This record does not assert independent reproduction.

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