# Adversarial Robust Image Classification on CIFAR-10 (test) (PGD20 & CW40)

Source: [Adversarial Robust Image Classification on CIFAR-10 (test) (PGD20 & CW40)](<https://www.sota2.com/research/sota/adversarial-robust-image-classification-on-cifar-10-test-pgd20-and-cw40>)

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

Data updated: 2026-02-26T06:02:23.239Z

Evaluation of Image Classification performance on the CIFAR-10 test set, focusing specifically on robustness metrics derived from PGD (20 iterations) and CW (40 iterations) adversarial attacks.

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 | Robust Accuracy (PGD-20) |
| 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

- [Robust Accuracy (PGD-20)](<https://www.sota2.com/research/sota/adversarial-robust-image-classification-on-cifar-10-test-pgd20-and-cw40.md?metric=Robust%20Accuracy%20(PGD-20)>): Higher is better; unit: PERCENT.

- [Robust Accuracy (CW-40)](<https://www.sota2.com/research/sota/adversarial-robust-image-classification-on-cifar-10-test-pgd20-and-cw40.md?metric=Robust%20Accuracy%20(CW-40)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Robust Accuracy (PGD-20) | Result date | Report paper |
| --- | --- | --- | --- |
| RobustResNet-A4 | 61.88 | Not recorded | 68fa6e3c1e209131cea3e8db |
| RobustResNet-A3 | 60.95 | Not recorded | 68fa6e3c1e209131cea3e8db |
| RobustResNet-A2 | 60.33 | Not recorded | 68fa6e3c1e209131cea3e8db |
| RobustResNet-A1 | 59.34 | Not recorded | 68fa6e3c1e209131cea3e8db |
| RobustResNet-A4 | 59.01 | Not recorded | 68fa6e3c1e209131cea3e8db |
| RobustResNet-A3 | 58.81 | Not recorded | 68fa6e3c1e209131cea3e8db |
| WRN-46-14 | 58.43 | Not recorded | 68fa6e3c1e209131cea3e8db |
| RobustResNet-A2 | 58.39 | Not recorded | 68fa6e3c1e209131cea3e8db |
| WRN-70-16 | 58.15 | Not recorded | 68fa6e3c1e209131cea3e8db |
| WRN-28-10 | 57.69 | Not recorded | 68fa6e3c1e209131cea3e8db |
| RobustResNet-A1 | 57.62 | Not recorded | 68fa6e3c1e209131cea3e8db |
| WRN-34-12 | 57.4 | Not recorded | 68fa6e3c1e209131cea3e8db |
| WRN-70-16 | 54.12 | Not recorded | 68fa6e3c1e209131cea3e8db |
| WRN-46-14 | 53.67 | Not recorded | 68fa6e3c1e209131cea3e8db |
| WRN-34-12 | 52.85 | Not recorded | 68fa6e3c1e209131cea3e8db |
| WRN-28-10 | 52.44 | Not recorded | 68fa6e3c1e209131cea3e8db |

### RobustResNet-A4

Robust Accuracy (PGD-20): 61.88. Result ID: 6990095d3a6c314849fda67b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Implementation](<https://github.com/zhichao-lu/robust-residual-network>)

Recorded evaluation configuration: {"#P (M)":"147","#F (G)":"39.4","Adversarial Training Protocol":"MART"}. This record does not assert independent reproduction.

### RobustResNet-A3

Robust Accuracy (PGD-20): 60.95. Result ID: 6990095d3a6c314849fda679.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Implementation](<https://github.com/zhichao-lu/robust-residual-network>)

Recorded evaluation configuration: {"#P (M)":"75.9","#F (G)":"19.9","Adversarial Training Protocol":"MART"}. This record does not assert independent reproduction.

### RobustResNet-A2

Robust Accuracy (PGD-20): 60.33. Result ID: 6990095d3a6c314849fda677.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Implementation](<https://github.com/zhichao-lu/robust-residual-network>)

Recorded evaluation configuration: {"#P (M)":"39.0","#F (G)":"10.8","Adversarial Training Protocol":"MART"}. This record does not assert independent reproduction.

### RobustResNet-A1

Robust Accuracy (PGD-20): 59.34. Result ID: 6990095d3a6c314849fda675.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Implementation](<https://github.com/zhichao-lu/robust-residual-network>)

Recorded evaluation configuration: {"#P (M)":"19.2","#F (G)":"5.11","Adversarial Training Protocol":"MART"}. This record does not assert independent reproduction.

### RobustResNet-A4

Robust Accuracy (PGD-20): 59.01. Result ID: 6990095d3a6c314849fda673.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Implementation](<https://github.com/zhichao-lu/robust-residual-network>)

Recorded evaluation configuration: {"#P (M)":"147","#F (G)":"39.4","Adversarial Training Protocol":"SAT"}. This record does not assert independent reproduction.

### RobustResNet-A3

Robust Accuracy (PGD-20): 58.81. Result ID: 6990095d3a6c314849fda671.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Implementation](<https://github.com/zhichao-lu/robust-residual-network>)

Recorded evaluation configuration: {"#P (M)":"75.9","#F (G)":"19.9","Adversarial Training Protocol":"SAT"}. This record does not assert independent reproduction.

### WRN-46-14

Robust Accuracy (PGD-20): 58.43. Result ID: 6990095d3a6c314849fda678.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63671e209131ce9fe70d.md>)

[Implementation](<https://github.com/szagoruyko/wide-residual-networks>)

Recorded evaluation configuration: {"#P (M)":"128","#F (G)":"18.6","Adversarial Training Protocol":"MART"}. This record does not assert independent reproduction.

### RobustResNet-A2

Robust Accuracy (PGD-20): 58.39. Result ID: 6990095d3a6c314849fda66f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Implementation](<https://github.com/zhichao-lu/robust-residual-network>)

Recorded evaluation configuration: {"#P (M)":"39.0","#F (G)":"10.8","Adversarial Training Protocol":"SAT"}. This record does not assert independent reproduction.

### WRN-70-16

Robust Accuracy (PGD-20): 58.15. Result ID: 6990095d3a6c314849fda67a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63671e209131ce9fe70d.md>)

[Implementation](<https://github.com/szagoruyko/wide-residual-networks>)

Recorded evaluation configuration: {"#P (M)":"267","#F (G)":"38.8","Adversarial Training Protocol":"MART"}. This record does not assert independent reproduction.

### WRN-28-10

Robust Accuracy (PGD-20): 57.69. Result ID: 6990095d3a6c314849fda674.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63671e209131ce9fe70d.md>)

[Implementation](<https://github.com/szagoruyko/wide-residual-networks>)

Recorded evaluation configuration: {"#P (M)":"36.5","#F (G)":"5.20","Adversarial Training Protocol":"MART"}. This record does not assert independent reproduction.

### RobustResNet-A1

Robust Accuracy (PGD-20): 57.62. Result ID: 6990095d3a6c314849fda66d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Implementation](<https://github.com/zhichao-lu/robust-residual-network>)

Recorded evaluation configuration: {"#P (M)":"19.2","#F (G)":"5.11","Adversarial Training Protocol":"SAT"}. This record does not assert independent reproduction.

### WRN-34-12

Robust Accuracy (PGD-20): 57.4. Result ID: 6990095d3a6c314849fda676.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63671e209131ce9fe70d.md>)

[Implementation](<https://github.com/szagoruyko/wide-residual-networks>)

Recorded evaluation configuration: {"#P (M)":"66.5","#F (G)":"9.60","Adversarial Training Protocol":"MART"}. This record does not assert independent reproduction.

### WRN-70-16

Robust Accuracy (PGD-20): 54.12. Result ID: 6990095d3a6c314849fda672.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63671e209131ce9fe70d.md>)

[Implementation](<https://github.com/szagoruyko/wide-residual-networks>)

Recorded evaluation configuration: {"#P (M)":"267","#F (G)":"38.8","Adversarial Training Protocol":"SAT"}. This record does not assert independent reproduction.

### WRN-46-14

Robust Accuracy (PGD-20): 53.67. Result ID: 6990095d3a6c314849fda670.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63671e209131ce9fe70d.md>)

[Implementation](<https://github.com/szagoruyko/wide-residual-networks>)

Recorded evaluation configuration: {"#P (M)":"128","#F (G)":"18.6","Adversarial Training Protocol":"SAT"}. This record does not assert independent reproduction.

### WRN-34-12

Robust Accuracy (PGD-20): 52.85. Result ID: 6990095d3a6c314849fda66e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63671e209131ce9fe70d.md>)

[Implementation](<https://github.com/szagoruyko/wide-residual-networks>)

Recorded evaluation configuration: {"#P (M)":"66.5","#F (G)":"9.60","Adversarial Training Protocol":"SAT"}. This record does not assert independent reproduction.

### WRN-28-10

Robust Accuracy (PGD-20): 52.44. Result ID: 6990095d3a6c314849fda66c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e3c1e209131cea3e8db.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63671e209131ce9fe70d.md>)

[Implementation](<https://github.com/szagoruyko/wide-residual-networks>)

Recorded evaluation configuration: {"#P (M)":"36.5","#F (G)":"5.20","Adversarial Training Protocol":"SAT"}. This record does not assert independent reproduction.

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