# Image Classification on CIFAR-10 (test) (PGD & Auto Attack Robustness)

Source: [Image Classification on CIFAR-10 (test) (PGD & Auto Attack Robustness)](<https://www.sota2.com/research/sota/image-classification-on-cifar-10-test-pgd-and-auto-attack-robustness>)

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

Data updated: 2026-02-26T05:34:20.852Z

Evaluation of Image Classification performance on the CIFAR-10 test set, focusing specifically on robustness metrics derived from Projected Gradient Descent (PGD) and general Auto Attack 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 | Image Classification |
| Dataset (including any stated split) | CIFAR-10 (test) |
| Metric | Accuracy (Natural) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 6 of 6 |
| 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

- [Accuracy (Natural)](<https://www.sota2.com/research/sota/image-classification-on-cifar-10-test-pgd-and-auto-attack-robustness.md?metric=Accuracy%20(Natural)>): Higher is better; unit: PERCENT.

- [Accuracy (PGD-20)](<https://www.sota2.com/research/sota/image-classification-on-cifar-10-test-pgd-and-auto-attack-robustness.md?metric=Accuracy%20(PGD-20)>): Higher is better; unit: PERCENT.

- [Accuracy (Auto Attack)](<https://www.sota2.com/research/sota/image-classification-on-cifar-10-test-pgd-and-auto-attack-robustness.md?metric=Accuracy%20(Auto%20Attack)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy (Natural) | Result date | Report paper |
| --- | --- | --- | --- |
| AUGMIX | 88.82 | Not recorded | 694cc4e953fdd5c36173a68a |
| Baseline | 87.77 | Not recorded | 694cc4e953fdd5c36173a68a |
| RandAugment | 84.63 | Not recorded | 694cc4e953fdd5c36173a68a |
| RandAugment + DOM\_DA | 84.52 | Not recorded | 694cc4e953fdd5c36173a68a |
| AUGMIX + DOM\_DA | 84.31 | Not recorded | 694cc4e953fdd5c36173a68a |
| Baseline + DOM\_RE | 71.66 | Not recorded | 694cc4e953fdd5c36173a68a |

### AUGMIX

Accuracy (Natural): 88.82. Result ID: 699941505698665ec4ab6219.

[Reporting paper](<https://www.sota2.com/research/paper/694cc4e953fdd5c36173a68a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13ccb.md>)

[Implementation](<https://github.com/google-research/augmix>)

Recorded evaluation configuration: {"Training Paradigm":"Single-step adversarial training","Checkpoint Type":"Final"}. This record does not assert independent reproduction.

### Baseline

Accuracy (Natural): 87.77. Result ID: 699941505698665ec4ab6217.

[Reporting paper](<https://www.sota2.com/research/paper/694cc4e953fdd5c36173a68a.md>)

Recorded evaluation configuration: {"Training Paradigm":"Single-step adversarial training","Checkpoint Type":"Final"}. This record does not assert independent reproduction.

### RandAugment

Accuracy (Natural): 84.63. Result ID: 699941505698665ec4ab621b.

[Reporting paper](<https://www.sota2.com/research/paper/694cc4e953fdd5c36173a68a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66af1e209131cea11c3c.md>)

[Implementation](<https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet>)

Recorded evaluation configuration: {"Training Paradigm":"Single-step adversarial training","Checkpoint Type":"Final"}. This record does not assert independent reproduction.

### RandAugment + DOM\_DA

Accuracy (Natural): 84.52. Result ID: 699941505698665ec4ab621c.

[Reporting paper](<https://www.sota2.com/research/paper/694cc4e953fdd5c36173a68a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/694cc4e953fdd5c36173a68a.md>)

Recorded evaluation configuration: {"Training Paradigm":"Single-step adversarial training","Checkpoint Type":"Final"}. This record does not assert independent reproduction.

### AUGMIX + DOM\_DA

Accuracy (Natural): 84.31. Result ID: 699941505698665ec4ab621a.

[Reporting paper](<https://www.sota2.com/research/paper/694cc4e953fdd5c36173a68a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/694cc4e953fdd5c36173a68a.md>)

Recorded evaluation configuration: {"Training Paradigm":"Single-step adversarial training","Checkpoint Type":"Final"}. This record does not assert independent reproduction.

### Baseline + DOM\_RE

Accuracy (Natural): 71.66. Result ID: 699941505698665ec4ab6218.

[Reporting paper](<https://www.sota2.com/research/paper/694cc4e953fdd5c36173a68a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/694cc4e953fdd5c36173a68a.md>)

Recorded evaluation configuration: {"Training Paradigm":"Single-step adversarial training","Checkpoint Type":"Final"}. This record does not assert independent reproduction.

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