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

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

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

Data updated: 2026-02-27T10:07:26.300Z

Evaluation of Image Classification performance on the validation split of the ImageNet dataset, specifically measuring robustness against Projected Gradient Descent (PGD) attacks with perturbation budgets of 10, 50, and 100 steps.

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 | Robust Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 10 of 10 |
| 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

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

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

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

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

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

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

## Recorded results

| Method | Robust Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| TTC | 38.41 | Not recorded | 68fa774a1e209131cea724cc |
| TTE | 31.01 | Not recorded | 68fa774a1e209131cea724cc |
| PMG-AFT | 21.43 | Not recorded | 68fa774a1e209131cea724cc |
| TeCoA | 18.89 | Not recorded | 68fa774a1e209131cea724cc |
| FARE | 14 | Not recorded | 68fa774a1e209131cea724cc |
| Anti-adv | 8.67 | Not recorded | 68fa774a1e209131cea724cc |
| HD | 6.63 | Not recorded | 68fa774a1e209131cea724cc |
| RN | 1.77 | Not recorded | 68fa774a1e209131cea724cc |
| CLIP | 1.15 | Not recorded | 68fa774a1e209131cea724cc |
| CLIP-FT | 0.93 | Not recorded | 68fa774a1e209131cea724cc |

### TTC

Robust Accuracy: 38.41. Result ID: 69a16cc635dfcabb00e6f5ae.

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

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

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

Recorded evaluation configuration: {"Mode":"Test-time Defence","Attack":"10-step PGD","epsilon":"1/255"}. This record does not assert independent reproduction.

### TTE

Robust Accuracy: 31.01. Result ID: 69a16cc635dfcabb00e6f5ab.

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

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a711e209131cea2817e.md>)

Recorded evaluation configuration: {"Mode":"Test-time Defence","Attack":"10-step PGD","epsilon":"1/255"}. This record does not assert independent reproduction.

### PMG-AFT

Robust Accuracy: 21.43. Result ID: 69a16cc635dfcabb00e6f5a8.

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

[Method origin paper](<https://www.sota2.com/research/paper/68fa72df1e209131cea5912e.md>)

Recorded evaluation configuration: {"Mode":"Adversarial Finetuning","Attack":"10-step PGD","epsilon":"1/255"}. This record does not assert independent reproduction.

### TeCoA

Robust Accuracy: 18.89. Result ID: 69a16cc635dfcabb00e6f5a7.

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

[Method origin paper](<https://www.sota2.com/research/paper/68fa6f271e209131cea44177.md>)

Recorded evaluation configuration: {"Mode":"Adversarial Finetuning","Attack":"10-step PGD","epsilon":"1/255"}. This record does not assert independent reproduction.

### FARE

Robust Accuracy: 14. Result ID: 69a16cc635dfcabb00e6f5a9.

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

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

Recorded evaluation configuration: {"Mode":"Adversarial Finetuning","Attack":"10-step PGD","epsilon":"1/255"}. This record does not assert independent reproduction.

### Anti-adv

Robust Accuracy: 8.67. Result ID: 69a16cc635dfcabb00e6f5ac.

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

[Method origin paper](<https://www.sota2.com/research/paper/68fa6b7f1e209131cea2e9af.md>)

Recorded evaluation configuration: {"Mode":"Test-time Defence","Attack":"10-step PGD","epsilon":"1/255"}. This record does not assert independent reproduction.

### HD

Robust Accuracy: 6.63. Result ID: 69a16cc635dfcabb00e6f5ad.

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

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a1d1e209131cea26045.md>)

Recorded evaluation configuration: {"Mode":"Test-time Defence","Attack":"10-step PGD","epsilon":"1/255"}. This record does not assert independent reproduction.

### RN

Robust Accuracy: 1.77. Result ID: 69a16cc635dfcabb00e6f5aa.

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

Recorded evaluation configuration: {"Mode":"Test-time Defence","Attack":"10-step PGD","epsilon":"1/255"}. This record does not assert independent reproduction.

### CLIP

Robust Accuracy: 1.15. Result ID: 69a16cc635dfcabb00e6f5a5.

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

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

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

Recorded evaluation configuration: {"Attack":"10-step PGD","epsilon":"1/255"}. This record does not assert independent reproduction.

### CLIP-FT

Robust Accuracy: 0.93. Result ID: 69a16cc635dfcabb00e6f5a6.

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

Recorded evaluation configuration: {"Mode":"Adversarial Finetuning","Attack":"10-step PGD","epsilon":"1/255"}. This record does not assert independent reproduction.

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