# Image Classification on ImageNet-1K (test val) (Backdoor Attack Metrics)

Source: [Image Classification on ImageNet-1K (test val) (Backdoor Attack Metrics)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-test-val-backdoor-attack-metrics>)

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

Data updated: 2026-02-27T10:58:07.147Z

Evaluation of Image Classification model robustness against backdoor attacks on the ImageNet-1K test/validation split, reporting metrics like Attack Success Rate (ASR) and Clean Accuracy (C-ACC).

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 val) |
| Metric | Clean Accuracy (C-ACC) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 5 of 5 |
| 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 (C-ACC)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-test-val-backdoor-attack-metrics.md?metric=Clean%20Accuracy%20(C-ACC)>): Higher is better; unit: PERCENT.

- [Backdoor Label Found](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-test-val-backdoor-attack-metrics.md?metric=Backdoor%20Label%20Found>): Higher is better; unit: SCALAR.

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

- [Delta Clean Accuracy (∆C-ACC)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-test-val-backdoor-attack-metrics.md?metric=Delta%20Clean%20Accuracy%20(%E2%88%86C-ACC)>): Higher is better; unit: SCALAR.

- [ASR2 Average](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-test-val-backdoor-attack-metrics.md?metric=ASR2%20Average>): Higher is better; unit: PERCENT.

- [ASR3](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-test-val-backdoor-attack-metrics.md?metric=ASR3>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Clean Accuracy (C-ACC) | Result date | Report paper |
| --- | --- | --- | --- |
| Fine-Pruning | 79.55 | Not recorded | 6982d2e2a8b01e76cc8e9873 |
| Baseline (no defense) | 78.88 | Not recorded | 6982d2e2a8b01e76cc8e9873 |
| BDVT | 78.58 | Not recorded | 6982d2e2a8b01e76cc8e9873 |
| Patch Processing | 76.07 | Not recorded | 6982d2e2a8b01e76cc8e9873 |

### Fine-Pruning

Clean Accuracy (C-ACC): 79.55. Result ID: 69a178b835dfcabb00e745cf.

[Reporting paper](<https://www.sota2.com/research/paper/6982d2e2a8b01e76cc8e9873.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa646b1e209131cea04983.md>)

Recorded evaluation configuration: {"Backbone":"DeiT-Small","Attack Configuration":"2-of-3"}. This record does not assert independent reproduction.

### Baseline (no defense)

Clean Accuracy (C-ACC): 78.88. Result ID: 69a178b835dfcabb00e745ce.

[Reporting paper](<https://www.sota2.com/research/paper/6982d2e2a8b01e76cc8e9873.md>)

Recorded evaluation configuration: {"Backbone":"DeiT-Small","Attack Configuration":"2-of-3"}. This record does not assert independent reproduction.

### BDVT

Clean Accuracy (C-ACC): 78.58. Result ID: 69a178b835dfcabb00e745d1.

[Reporting paper](<https://www.sota2.com/research/paper/6982d2e2a8b01e76cc8e9873.md>)

Recorded evaluation configuration: {"Backbone":"DeiT-Small","Attack Configuration":"2-of-3"}. This record does not assert independent reproduction.

### Patch Processing

Clean Accuracy (C-ACC): 76.07. Result ID: 69a178b835dfcabb00e745d0.

[Reporting paper](<https://www.sota2.com/research/paper/6982d2e2a8b01e76cc8e9873.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e5b1e209131cea3f4ff.md>)

Recorded evaluation configuration: {"Backbone":"DeiT-Small","Attack Configuration":"2-of-3"}. This record does not assert independent reproduction.

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