# Image Classification on TinyImageNet (Adversarial Robustness Suite)

Source: [Image Classification on TinyImageNet (Adversarial Robustness Suite)](<https://www.sota2.com/research/sota/image-classification-on-tinyimagenet-adversarial-robustness-suite>)

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

Data updated: 2026-02-27T11:46:09.213Z

Evaluation of Image Classification performance on the TinyImageNet dataset, measuring robustness against a specific suite of adversarial attacks including PGD, CW, and AutoAttack, 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) | TinyImageNet |
| Metric | Clean Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 17 of 17 |
| 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-tinyimagenet-adversarial-robustness-suite.md?metric=Clean%20Accuracy>): Higher is better; unit: PERCENT.

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

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

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

## Recorded results

| Method | Clean Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| TGA-ZSR | 76.6 | Not recorded | 6940e7a8456146936284f897 |
| UCAT | 74.46 | Not recorded | 6940e7a8456146936284f897 |
| UCAT | 74.46 | Not recorded | 6940e7a8456146936284f897 |
| ACAT | 72.8 | Not recorded | 6940e7a8456146936284f897 |
| TeCoA | 71.24 | Not recorded | 6940e7a8456146936284f897 |
| DKL | 70.84 | Not recorded | 6940e7a8456146936284f897 |
| TGA-ZSR | 69.78 | Not recorded | 6940e7a8456146936284f897 |
| TRADES | 67.73 | Not recorded | 6940e7a8456146936284f897 |
| UCAT | 67.18 | Not recorded | 6940e7a8456146936284f897 |
| TeCoA | 63.2 | Not recorded | 6940e7a8456146936284f897 |
| CLIP | 57.96 | Not recorded | 6940e7a8456146936284f897 |
| CLIP | 57.96 | Not recorded | 6940e7a8456146936284f897 |
| CLIP | 57.96 | Not recorded | 6940e7a8456146936284f897 |
| PMG-AFT | 48.6 | Not recorded | 6940e7a8456146936284f897 |
| FARE | 41.86 | Not recorded | 6940e7a8456146936284f897 |
| PMG-AFT | 22.16 | Not recorded | 6940e7a8456146936284f897 |
| FARE | 16.92 | Not recorded | 6940e7a8456146936284f897 |

### TGA-ZSR

Clean Accuracy: 76.6. Result ID: 69a183f135dfcabb00e788fb.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75771e209131cea68256.md>)

[Implementation](<https://github.com/zhyblue424/TGA-ZSR>)

Recorded evaluation configuration: {"Fine-tuning dataset":"TinyImageNet","Adversarial training":"2-step PGD","Epsilon":"1/255"}. This record does not assert independent reproduction.

### UCAT

Clean Accuracy: 74.46. Result ID: 69a183f135dfcabb00e7884c.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

Recorded evaluation configuration: {"Fine-tuning dataset":"TinyImageNet","Adversarial training":"2-step PGD","Epsilon":"1/255","Evaluation protocol":"Zero-shot"}. This record does not assert independent reproduction.

### UCAT

Clean Accuracy: 74.46. Result ID: 69a183f135dfcabb00e788fc.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

Recorded evaluation configuration: {"Fine-tuning dataset":"TinyImageNet","Adversarial training":"2-step PGD","Epsilon":"1/255"}. This record does not assert independent reproduction.

### ACAT

Clean Accuracy: 72.8. Result ID: 69a183f135dfcabb00e7884a.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6dc61e209131cea3c0ff.md>)

[Implementation](<https://github.com/val-iisc/DAJAT>)

Recorded evaluation configuration: {"Fine-tuning dataset":"TinyImageNet","Adversarial training":"2-step PGD","Epsilon":"1/255","Evaluation protocol":"Zero-shot"}. This record does not assert independent reproduction.

### TeCoA

Clean Accuracy: 71.24. Result ID: 69a183f135dfcabb00e788f8.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

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

Recorded evaluation configuration: {"Fine-tuning dataset":"TinyImageNet","Adversarial training":"2-step PGD","Epsilon":"1/255"}. This record does not assert independent reproduction.

### DKL

Clean Accuracy: 70.84. Result ID: 69a183f135dfcabb00e7884b.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa756d1e209131cea67d56.md>)

[Implementation](<https://github.com/jiequancui/DKL>)

Recorded evaluation configuration: {"Fine-tuning dataset":"TinyImageNet","Adversarial training":"2-step PGD","Epsilon":"1/255","Evaluation protocol":"Zero-shot"}. This record does not assert independent reproduction.

### TGA-ZSR

Clean Accuracy: 69.78. Result ID: 69a183f135dfcabb00e7896d.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75771e209131cea68256.md>)

[Implementation](<https://github.com/zhyblue424/TGA-ZSR>)

Recorded evaluation configuration: {"Training Dataset":"TinyImageNet","Adversarial Training":"10-step PGD","Training Epsilon":"2/255","Evaluation Protocol":"Zero-shot"}. This record does not assert independent reproduction.

### TRADES

Clean Accuracy: 67.73. Result ID: 69a183f135dfcabb00e78849.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65f71e209131cea0d897.md>)

Recorded evaluation configuration: {"Fine-tuning dataset":"TinyImageNet","Adversarial training":"2-step PGD","Epsilon":"1/255","Evaluation protocol":"Zero-shot"}. This record does not assert independent reproduction.

### UCAT

Clean Accuracy: 67.18. Result ID: 69a183f135dfcabb00e7896e.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

Recorded evaluation configuration: {"Training Dataset":"TinyImageNet","Adversarial Training":"10-step PGD","Training Epsilon":"2/255","Evaluation Protocol":"Zero-shot"}. This record does not assert independent reproduction.

### TeCoA

Clean Accuracy: 63.2. Result ID: 69a183f135dfcabb00e7896a.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

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

Recorded evaluation configuration: {"Training Dataset":"TinyImageNet","Adversarial Training":"10-step PGD","Training Epsilon":"2/255","Evaluation Protocol":"Zero-shot"}. This record does not assert independent reproduction.

### CLIP

Clean Accuracy: 57.96. Result ID: 69a183f135dfcabb00e78848.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

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

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

Recorded evaluation configuration: {"Fine-tuning dataset":"TinyImageNet","Adversarial training":"2-step PGD","Epsilon":"1/255","Evaluation protocol":"Zero-shot"}. This record does not assert independent reproduction.

### CLIP

Clean Accuracy: 57.96. Result ID: 69a183f135dfcabb00e788f7.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

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

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

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### CLIP

Clean Accuracy: 57.96. Result ID: 69a183f135dfcabb00e78969.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

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

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

Recorded evaluation configuration: {"Training Dataset":"ImageNet-21K / Pre-trained","Evaluation Protocol":"Zero-shot"}. This record does not assert independent reproduction.

### PMG-AFT

Clean Accuracy: 48.6. Result ID: 69a183f135dfcabb00e788fa.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

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

Recorded evaluation configuration: {"Fine-tuning dataset":"TinyImageNet","Adversarial training":"2-step PGD","Epsilon":"1/255"}. This record does not assert independent reproduction.

### FARE

Clean Accuracy: 41.86. Result ID: 69a183f135dfcabb00e788f9.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

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

Recorded evaluation configuration: {"Fine-tuning dataset":"TinyImageNet","Adversarial training":"2-step PGD","Epsilon":"1/255"}. This record does not assert independent reproduction.

### PMG-AFT

Clean Accuracy: 22.16. Result ID: 69a183f135dfcabb00e7896c.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

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

Recorded evaluation configuration: {"Training Dataset":"TinyImageNet","Adversarial Training":"10-step PGD","Training Epsilon":"2/255","Evaluation Protocol":"Zero-shot"}. This record does not assert independent reproduction.

### FARE

Clean Accuracy: 16.92. Result ID: 69a183f135dfcabb00e7896b.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7a8456146936284f897.md>)

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

Recorded evaluation configuration: {"Training Dataset":"TinyImageNet","Adversarial Training":"10-step PGD","Training Epsilon":"2/255","Evaluation Protocol":"Zero-shot"}. This record does not assert independent reproduction.

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