# Image Classification on ImageNet 1k (test) (Robustness Metrics)

Source: [Image Classification on ImageNet 1k (test) (Robustness Metrics)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-test-robustness-metrics>)

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

Data updated: 2026-02-27T11:07:15.204Z

Evaluation of Image Classification performance on the test split of the ImageNet 1k dataset, specifically reporting Clean Accuracy and Robust 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 1k (test) |
| Metric | Clean Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 14 of 14 |
| 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-imagenet-1k-test-robustness-metrics.md?metric=Clean%20Accuracy>): Higher is better; unit: PERCENT.

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

## Recorded results

| Method | Clean Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| ShapePuri | 84.06 | Not recorded | 69857571c97a0b3e3b5d5b64 |
| Without defense | 82.33 | Not recorded | 69857571c97a0b3e3b5d5b64 |
| ShapePuri | 80.73 | Not recorded | 69857571c97a0b3e3b5d5b64 |
| Without defense | 80.55 | Not recorded | 69857571c97a0b3e3b5d5b64 |
| Without defense | 80.55 | Not recorded | 69857571c97a0b3e3b5d5b64 |
| OSCP | 79.81 | Not recorded | 69857571c97a0b3e3b5d5b64 |
| ShapePuri | 79.47 | Not recorded | 69857571c97a0b3e3b5d5b64 |
| GDMP | 78.1 | Not recorded | 69857571c97a0b3e3b5d5b64 |
| MeanSparse | 77.96 | Not recorded | 69857571c97a0b3e3b5d5b64 |
| OSCP | 77.63 | Not recorded | 69857571c97a0b3e3b5d5b64 |
| OSCP | 77.63 | Not recorded | 69857571c97a0b3e3b5d5b64 |
| Singh et al. | 77 | Not recorded | 69857571c97a0b3e3b5d5b64 |
| DiffPure | 75.77 | Not recorded | 69857571c97a0b3e3b5d5b64 |
| GDMP | 73.53 | Not recorded | 69857571c97a0b3e3b5d5b64 |

### ShapePuri

Clean Accuracy: 84.06. Result ID: 69a17ad935dfcabb00e756a3.

[Reporting paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

Recorded evaluation configuration: {"Attacks":"AutoAttack","Classifiers":"ConvNeXt-L","Domain":"Hybrid"}. This record does not assert independent reproduction.

### Without defense

Clean Accuracy: 82.33. Result ID: 69a17ad935dfcabb00e7569a.

[Reporting paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

Recorded evaluation configuration: {"Attacks":"targeted PGD-40","Classifiers":"ResNet-152","Domain":"Without defense"}. This record does not assert independent reproduction.

### ShapePuri

Clean Accuracy: 80.73. Result ID: 69a17ad935dfcabb00e7569d.

[Reporting paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

Recorded evaluation configuration: {"Attacks":"targeted PGD-40","Classifiers":"ResNet-152","Domain":"Hybrid"}. This record does not assert independent reproduction.

### Without defense

Clean Accuracy: 80.55. Result ID: 69a17ad935dfcabb00e75696.

[Reporting paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

Recorded evaluation configuration: {"Attacks":"untargeted PGD-100","Classifiers":"ResNet-50","Domain":"Without defense"}. This record does not assert independent reproduction.

### Without defense

Clean Accuracy: 80.55. Result ID: 69a17ad935dfcabb00e7569e.

[Reporting paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

Recorded evaluation configuration: {"Attacks":"AutoAttack","Classifiers":"ResNet-50","Domain":"Without defense"}. This record does not assert independent reproduction.

### OSCP

Clean Accuracy: 79.81. Result ID: 69a17ad935dfcabb00e7569c.

[Reporting paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77991e209131cea74467.md>)

Recorded evaluation configuration: {"Attacks":"targeted PGD-40","Classifiers":"ResNet-152","Domain":"Hybrid"}. This record does not assert independent reproduction.

### ShapePuri

Clean Accuracy: 79.47. Result ID: 69a17ad935dfcabb00e75699.

[Reporting paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

Recorded evaluation configuration: {"Attacks":"untargeted PGD-100","Classifiers":"ResNet-101","Domain":"Hybrid"}. This record does not assert independent reproduction.

### GDMP

Clean Accuracy: 78.1. Result ID: 69a17ad935dfcabb00e7569b.

[Reporting paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ce61e209131cea369e5.md>)

Recorded evaluation configuration: {"Attacks":"targeted PGD-40","Classifiers":"ResNet-152","Domain":"Adversarial Purification"}. This record does not assert independent reproduction.

### MeanSparse

Clean Accuracy: 77.96. Result ID: 69a17ad935dfcabb00e7569f.

[Reporting paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7bfc1e209131cea8ba6c.md>)

Recorded evaluation configuration: {"Attacks":"AutoAttack","Classifiers":"ConvNeXt-L","Domain":"Adversarial Training"}. This record does not assert independent reproduction.

### OSCP

Clean Accuracy: 77.63. Result ID: 69a17ad935dfcabb00e75698.

[Reporting paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77991e209131cea74467.md>)

Recorded evaluation configuration: {"Attacks":"untargeted PGD-100","Classifiers":"ResNet-50","Domain":"Hybrid"}. This record does not assert independent reproduction.

### OSCP

Clean Accuracy: 77.63. Result ID: 69a17ad935dfcabb00e756a2.

[Reporting paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77991e209131cea74467.md>)

Recorded evaluation configuration: {"Attacks":"AutoAttack","Classifiers":"ResNet-50","Domain":"Hybrid"}. This record does not assert independent reproduction.

### Singh et al.

Clean Accuracy: 77. Result ID: 69a17ad935dfcabb00e756a0.

[Reporting paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71011e209131cea4e7c0.md>)

[Implementation](<https://github.com/nmndeep/revisiting-at>)

Recorded evaluation configuration: {"Attacks":"AutoAttack","Classifiers":"ConvNeXt-L","Domain":"Adversarial Training"}. This record does not assert independent reproduction.

### DiffPure

Clean Accuracy: 75.77. Result ID: 69a17ad935dfcabb00e756a1.

[Reporting paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c841e209131cea3472d.md>)

Recorded evaluation configuration: {"Attacks":"AutoAttack","Classifiers":"ResNet-50","Domain":"Adversarial Purification"}. This record does not assert independent reproduction.

### GDMP

Clean Accuracy: 73.53. Result ID: 69a17ad935dfcabb00e75697.

[Reporting paper](<https://www.sota2.com/research/paper/69857571c97a0b3e3b5d5b64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ce61e209131cea369e5.md>)

Recorded evaluation configuration: {"Attacks":"untargeted PGD-100","Classifiers":"ResNet-50","Domain":"Adversarial Purification"}. This record does not assert independent reproduction.

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