# Image Classification on ImageNet-A (Adversarial Robustness)

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

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

Data updated: 2026-05-20T09:29:21.910Z

Evaluation of Image Classification performance on the ImageNet-A dataset, specifically measuring robustness against adversarial examples alongside clean 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-A |
| Metric | Top-1 Accuracy (Clean) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 16 of 16 |
| 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

- [Top-1 Accuracy (Clean)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-a-adversarial-robustness.md?metric=Top-1%20Accuracy%20(Clean)>): Higher is better; unit: PERCENT.

- [Top-1 Accuracy (Adversarial)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-a-adversarial-robustness.md?metric=Top-1%20Accuracy%20(Adversarial)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Top-1 Accuracy (Clean) | Result date | Report paper |
| --- | --- | --- | --- |
| CLIP | 32.69 | Not recorded | 69cb54ee0636e5bf4afc9d9a |
| WiSE-FT | 32.6 | Not recorded | 69cb54ee0636e5bf4afc9d9a |
| TPGM | 32.38 | Not recorded | 69cb54ee0636e5bf4afc9d9a |
| SPD | 32.07 | Not recorded | 69cb54ee0636e5bf4afc9d9a |
| A-TPT | 31.6 | Not recorded | 6a0d4031f984739ed673796d |
| FLYP | 30.06 | Not recorded | 69cb54ee0636e5bf4afc9d9a |
| R-TPT | 28.4 | Not recorded | 6a0d4031f984739ed673796d |
| MTA | 27.5 | Not recorded | 6a0d4031f984739ed673796d |
| TPT-Ensemble | 22.6 | Not recorded | 6a0d4031f984739ed673796d |
| Vanilla FT | 22.47 | Not recorded | 69cb54ee0636e5bf4afc9d9a |
| CLIP | 21.8 | Not recorded | 6a0d4031f984739ed673796d |
| GRACE | 20.02 | Not recorded | 69cb54ee0636e5bf4afc9d9a |
| LAAT | 12.23 | Not recorded | 69cb54ee0636e5bf4afc9d9a |
| PMG-AFT | 8.22 | Not recorded | 69cb54ee0636e5bf4afc9d9a |
| FARE | 5.97 | Not recorded | 69cb54ee0636e5bf4afc9d9a |
| TeCoA | 5.11 | Not recorded | 69cb54ee0636e5bf4afc9d9a |

### CLIP

Top-1 Accuracy (Clean): 32.69. Result ID: 69cb824347f34ce924b3b243.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54ee0636e5bf4afc9d9a.md>)

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

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

Recorded evaluation configuration: {"Backbone":"ViT-B/32"}. This record does not assert independent reproduction.

### WiSE-FT

Top-1 Accuracy (Clean): 32.6. Result ID: 69cb824347f34ce924b3b245.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54ee0636e5bf4afc9d9a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6cd31e209131cea3601a.md>)

[Implementation](<https://github.com/mlfoundations/wise-ft>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32"}. This record does not assert independent reproduction.

### TPGM

Top-1 Accuracy (Clean): 32.38. Result ID: 69cb824347f34ce924b3b247.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54ee0636e5bf4afc9d9a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6edf1e209131cea428cb.md>)

[Implementation](<https://github.com/PotatoTian/TPGM>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32"}. This record does not assert independent reproduction.

### SPD

Top-1 Accuracy (Clean): 32.07. Result ID: 69cb824347f34ce924b3b248.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54ee0636e5bf4afc9d9a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75841e209131cea6855b.md>)

[Implementation](<https://github.com/GT-RIPL/Selective-Projection-Decay.git>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32"}. This record does not assert independent reproduction.

### A-TPT

Top-1 Accuracy (Clean): 31.6. Result ID: 6a0d7edefbc373592c05dc82.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4031f984739ed673796d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0d4031f984739ed673796d.md>)

Recorded evaluation configuration: {"Backbone":"ResNet50"}. This record does not assert independent reproduction.

### FLYP

Top-1 Accuracy (Clean): 30.06. Result ID: 69cb824347f34ce924b3b246.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54ee0636e5bf4afc9d9a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e241e209131cea3e1b0.md>)

[Implementation](<https://github.com/locuslab/FLYP>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32"}. This record does not assert independent reproduction.

### R-TPT

Top-1 Accuracy (Clean): 28.4. Result ID: 6a0d7edefbc373592c05dc81.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4031f984739ed673796d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7add1e209131cea8590d.md>)

[Implementation](<https://github.com/TomSheng21/R-TPT>)

Recorded evaluation configuration: {"Backbone":"ResNet50"}. This record does not assert independent reproduction.

### MTA

Top-1 Accuracy (Clean): 27.5. Result ID: 6a0d7edefbc373592c05dc80.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4031f984739ed673796d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73191e209131cea5a896.md>)

[Implementation](<https://github.com/MaxZanella/MTA>)

Recorded evaluation configuration: {"Backbone":"ResNet50"}. This record does not assert independent reproduction.

### TPT-Ensemble

Top-1 Accuracy (Clean): 22.6. Result ID: 6a0d7edefbc373592c05dc7f.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4031f984739ed673796d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d691e209131cea399f9.md>)

[Implementation](<https://tinyurl.com/yr3zmhma>)

Recorded evaluation configuration: {"Backbone":"ResNet50"}. This record does not assert independent reproduction.

### Vanilla FT

Top-1 Accuracy (Clean): 22.47. Result ID: 69cb824347f34ce924b3b244.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54ee0636e5bf4afc9d9a.md>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32"}. This record does not assert independent reproduction.

### CLIP

Top-1 Accuracy (Clean): 21.8. Result ID: 6a0d7edefbc373592c05dc7e.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4031f984739ed673796d.md>)

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

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

Recorded evaluation configuration: {"Backbone":"ResNet50"}. This record does not assert independent reproduction.

### GRACE

Top-1 Accuracy (Clean): 20.02. Result ID: 69cb824347f34ce924b3b24d.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54ee0636e5bf4afc9d9a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69cb54ee0636e5bf4afc9d9a.md>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32"}. This record does not assert independent reproduction.

### LAAT

Top-1 Accuracy (Clean): 12.23. Result ID: 69cb824347f34ce924b3b24c.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54ee0636e5bf4afc9d9a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72701e209131cea5651c.md>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32"}. This record does not assert independent reproduction.

### PMG-AFT

Top-1 Accuracy (Clean): 8.22. Result ID: 69cb824347f34ce924b3b24b.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54ee0636e5bf4afc9d9a.md>)

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

Recorded evaluation configuration: {"Backbone":"ViT-B/32"}. This record does not assert independent reproduction.

### FARE

Top-1 Accuracy (Clean): 5.97. Result ID: 69cb824347f34ce924b3b24a.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54ee0636e5bf4afc9d9a.md>)

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

Recorded evaluation configuration: {"Backbone":"ViT-B/32"}. This record does not assert independent reproduction.

### TeCoA

Top-1 Accuracy (Clean): 5.11. Result ID: 69cb824347f34ce924b3b249.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54ee0636e5bf4afc9d9a.md>)

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

Recorded evaluation configuration: {"Backbone":"ViT-B/32"}. This record does not assert independent reproduction.

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