# Image Classification on ImageNet (Robustness & Calibration Evaluation)

Source: [Image Classification on ImageNet (Robustness & Calibration Evaluation)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-robustness-and-calibration-evaluation>)

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

Data updated: 2026-02-26T05:41:21.104Z

Evaluation of Image Classification performance on the standard ImageNet dataset, focusing on model calibration (ECE) and robustness against various query-based adversarial attacks (e.g., Square, SignHunter, SimBA).

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 |
| Metric | ECE |
| Unit | PERCENT |
| Direction | Lower is better |
| Results loaded | 8 of 8 |
| 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

- [ECE](<https://www.sota2.com/research/sota/image-classification-on-imagenet-robustness-and-calibration-evaluation.md?metric=ECE>): Lower is better; unit: PERCENT.

- [Accuracy](<https://www.sota2.com/research/sota/image-classification-on-imagenet-robustness-and-calibration-evaluation.md?metric=Accuracy>): Higher is better; unit: PERCENT.

- [Robust Acc (Square, 100 queries)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-robustness-and-calibration-evaluation.md?metric=Robust%20Acc%20(Square%2C%20100%20queries)>): Higher is better; unit: PERCENT.

- [Robust Acc (Square, 2500 queries)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-robustness-and-calibration-evaluation.md?metric=Robust%20Acc%20(Square%2C%202500%20queries)>): Higher is better; unit: PERCENT.

- [Robust Acc (SignHunter, 100 queries)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-robustness-and-calibration-evaluation.md?metric=Robust%20Acc%20(SignHunter%2C%20100%20queries)>): Higher is better; unit: PERCENT.

- [Robust Acc (SignHunter, 2500 queries)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-robustness-and-calibration-evaluation.md?metric=Robust%20Acc%20(SignHunter%2C%202500%20queries)>): Higher is better; unit: PERCENT.

- [Robust Acc (SimBA, 100 queries)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-robustness-and-calibration-evaluation.md?metric=Robust%20Acc%20(SimBA%2C%20100%20queries)>): Higher is better; unit: PERCENT.

- [Robust Acc (SimBA, 2500 queries)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-robustness-and-calibration-evaluation.md?metric=Robust%20Acc%20(SimBA%2C%202500%20queries)>): Higher is better; unit: PERCENT.

- [Robustness Score (NES, 100 queries)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-robustness-and-calibration-evaluation.md?metric=Robustness%20Score%20(NES%2C%20100%20queries)>): Higher is better; unit: PERCENT.

- [Robustness Score (NES, 2500 queries)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-robustness-and-calibration-evaluation.md?metric=Robustness%20Score%20(NES%2C%202500%20queries)>): Higher is better; unit: PERCENT.

- [Robust Acc (Bandit, 100 queries)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-robustness-and-calibration-evaluation.md?metric=Robust%20Acc%20(Bandit%2C%20100%20queries)>): Higher is better; unit: PERCENT.

- [Robust Acc (Bandit, 2500 queries)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-robustness-and-calibration-evaluation.md?metric=Robust%20Acc%20(Bandit%2C%202500%20queries)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | ECE | Result date | Report paper |
| --- | --- | --- | --- |
| AAA-linear | 4.3 | Not recorded | 68fa6e401e209131cea3eb22 |
| AT | 5.03 | Not recorded | 68fa6e401e209131cea3eb22 |
| None | 5.42 | Not recorded | 68fa6e401e209131cea3eb22 |
| AT | 5.74 | Not recorded | 68fa6e401e209131cea3eb22 |
| RND | 5.79 | Not recorded | 68fa6e401e209131cea3eb22 |
| AAA-linear | 7.38 | Not recorded | 68fa6e401e209131cea3eb22 |
| None | 8.37 | Not recorded | 68fa6e401e209131cea3eb22 |
| RND | 8.93 | Not recorded | 68fa6e401e209131cea3eb22 |

### AAA-linear

ECE: 4.3. Result ID: 69901d733a6c314849fe3e54.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e401e209131cea3eb22.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e401e209131cea3eb22.md>)

[Implementation](<https://github.com/Sizhe-Chen/AAA>)

Recorded evaluation configuration: {"Backbone":"Wide-ResNet-50","Perturbation Budget (l\_inf)":"4/255"}. This record does not assert independent reproduction.

### AT

ECE: 5.03. Result ID: 69901d733a6c314849fe3e52.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e401e209131cea3eb22.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6af61e209131cea2b5f7.md>)

Recorded evaluation configuration: {"Backbone":"Wide-ResNet-50","Perturbation Budget (l\_inf)":"4/255"}. This record does not assert independent reproduction.

### None

ECE: 5.42. Result ID: 69901d733a6c314849fe3e51.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e401e209131cea3eb22.md>)

Recorded evaluation configuration: {"Backbone":"Wide-ResNet-50","Perturbation Budget (l\_inf)":"4/255"}. This record does not assert independent reproduction.

### AT

ECE: 5.74. Result ID: 69901d733a6c314849fe3e56.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e401e209131cea3eb22.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6af61e209131cea2b5f7.md>)

Recorded evaluation configuration: {"Backbone":"ResNeXt-101","Perturbation Budget (l\_inf)":"4/255"}. This record does not assert independent reproduction.

### RND

ECE: 5.79. Result ID: 69901d733a6c314849fe3e53.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e401e209131cea3eb22.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6b2d1e209131cea2c9d5.md>)

Recorded evaluation configuration: {"Backbone":"Wide-ResNet-50","Perturbation Budget (l\_inf)":"4/255"}. This record does not assert independent reproduction.

### AAA-linear

ECE: 7.38. Result ID: 69901d733a6c314849fe3e58.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e401e209131cea3eb22.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e401e209131cea3eb22.md>)

[Implementation](<https://github.com/Sizhe-Chen/AAA>)

Recorded evaluation configuration: {"Backbone":"ResNeXt-101","Perturbation Budget (l\_inf)":"4/255"}. This record does not assert independent reproduction.

### None

ECE: 8.37. Result ID: 69901d733a6c314849fe3e55.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e401e209131cea3eb22.md>)

Recorded evaluation configuration: {"Backbone":"ResNeXt-101","Perturbation Budget (l\_inf)":"4/255"}. This record does not assert independent reproduction.

### RND

ECE: 8.93. Result ID: 69901d733a6c314849fe3e57.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e401e209131cea3eb22.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6b2d1e209131cea2c9d5.md>)

Recorded evaluation configuration: {"Backbone":"ResNeXt-101","Perturbation Budget (l\_inf)":"4/255"}. This record does not assert independent reproduction.

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