# Backdoor Detection on Balanced 50% clean, 50% backdoored (test)

Source: [Backdoor Detection on Balanced 50% clean, 50% backdoored (test)](<https://www.sota2.com/research/sota/backdoor-detection-on-balanced-50-percent-clean-50-percent-backdoored-test>)

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

Data updated: 2026-02-26T05:40:46.198Z

Evaluation of Backdoor Detection performance on a dataset specifically constructed with a 50% clean and 50% backdoored split, measured by Detection 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 | Backdoor Detection |
| Dataset (including any stated split) | Balanced 50% clean, 50% backdoored (test) |
| Metric | Detection Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 28 of 28 |
| 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

- [Detection Accuracy](<https://www.sota2.com/research/sota/backdoor-detection-on-balanced-50-percent-clean-50-percent-backdoored-test.md?metric=Detection%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Detection Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| VillanDiffusion | 98.2 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| Personalization | 92.8 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| BadT2I | 90.8 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| Rickrolling | 88.3 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| VillanDiffusion | 87.4 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| VillanDiffusion | 78.9 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| VillanDiffusion | 75 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| Rickrolling | 64.5 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| SemBD | 57 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| EvilEdit | 55.5 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| BadT2I | 54.9 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| Rickrolling | 54.6 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| EvilEdit | 54.3 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| EvilEdit | 54 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| SemBD | 52.5 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| BadT2I | 52 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| BadT2I | 51.3 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| Personalization | 50.7 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| IBA | 49.5 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| IBA | 49.5 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| SemBD | 48.5 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| Rickrolling | 47 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| Personalization | 45.8 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| EvilEdit | 45.5 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| Personalization | 43 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| IBA | 42.5 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| IBA | 41 | Not recorded | 698575a0c97a0b3e3b5d5cab |
| SemBD | 39.5 | Not recorded | 698575a0c97a0b3e3b5d5cab |

### VillanDiffusion

Detection Accuracy: 98.2. Result ID: 6996dee35698665ec42b567d.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71a41e209131cea523f5.md>)

[Implementation](<https://github.com/IBM/villandiffusion>)

Recorded evaluation configuration: {"Defense Method":"NaviT2I"}. This record does not assert independent reproduction.

### Personalization

Detection Accuracy: 92.8. Result ID: 6996dee35698665ec42b5681.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71911e209131cea5206b.md>)

Recorded evaluation configuration: {"Defense Method":"NaviT2I"}. This record does not assert independent reproduction.

### BadT2I

Detection Accuracy: 90.8. Result ID: 6996dee35698665ec42b568d.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70f21e209131cea4e442.md>)

Recorded evaluation configuration: {"Defense Method":"NaviT2I"}. This record does not assert independent reproduction.

### Rickrolling

Detection Accuracy: 88.3. Result ID: 6996dee35698665ec42b5685.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70471e209131cea4a212.md>)

Recorded evaluation configuration: {"Defense Method":"NaviT2I"}. This record does not assert independent reproduction.

### VillanDiffusion

Detection Accuracy: 87.4. Result ID: 6996dee35698665ec42b567e.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71a41e209131cea523f5.md>)

[Implementation](<https://github.com/IBM/villandiffusion>)

Recorded evaluation configuration: {"Defense Method":"UFID"}. This record does not assert independent reproduction.

### VillanDiffusion

Detection Accuracy: 78.9. Result ID: 6996dee35698665ec42b5680.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71a41e209131cea523f5.md>)

[Implementation](<https://github.com/IBM/villandiffusion>)

Recorded evaluation configuration: {"Defense Method":"T2IShield\_CDA"}. This record does not assert independent reproduction.

### VillanDiffusion

Detection Accuracy: 75. Result ID: 6996dee35698665ec42b567f.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71a41e209131cea523f5.md>)

[Implementation](<https://github.com/IBM/villandiffusion>)

Recorded evaluation configuration: {"Defense Method":"T2IShield\_FFT"}. This record does not assert independent reproduction.

### Rickrolling

Detection Accuracy: 64.5. Result ID: 6996dee35698665ec42b5688.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70471e209131cea4a212.md>)

Recorded evaluation configuration: {"Defense Method":"T2IShield\_CDA"}. This record does not assert independent reproduction.

### SemBD

Detection Accuracy: 57. Result ID: 6996dee35698665ec42b5697.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

Recorded evaluation configuration: {"Defense Method":"T2IShield\_FFT"}. This record does not assert independent reproduction.

### EvilEdit

Detection Accuracy: 55.5. Result ID: 6996dee35698665ec42b568b.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

Recorded evaluation configuration: {"Defense Method":"T2IShield\_FFT"}. This record does not assert independent reproduction.

### BadT2I

Detection Accuracy: 54.9. Result ID: 6996dee35698665ec42b568e.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70f21e209131cea4e442.md>)

Recorded evaluation configuration: {"Defense Method":"UFID"}. This record does not assert independent reproduction.

### Rickrolling

Detection Accuracy: 54.6. Result ID: 6996dee35698665ec42b5686.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70471e209131cea4a212.md>)

Recorded evaluation configuration: {"Defense Method":"UFID"}. This record does not assert independent reproduction.

### EvilEdit

Detection Accuracy: 54.3. Result ID: 6996dee35698665ec42b5689.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

Recorded evaluation configuration: {"Defense Method":"NaviT2I"}. This record does not assert independent reproduction.

### EvilEdit

Detection Accuracy: 54. Result ID: 6996dee35698665ec42b568c.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

Recorded evaluation configuration: {"Defense Method":"T2IShield\_CDA"}. This record does not assert independent reproduction.

### SemBD

Detection Accuracy: 52.5. Result ID: 6996dee35698665ec42b5695.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

Recorded evaluation configuration: {"Defense Method":"NaviT2I"}. This record does not assert independent reproduction.

### BadT2I

Detection Accuracy: 52. Result ID: 6996dee35698665ec42b568f.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70f21e209131cea4e442.md>)

Recorded evaluation configuration: {"Defense Method":"T2IShield\_FFT"}. This record does not assert independent reproduction.

### BadT2I

Detection Accuracy: 51.3. Result ID: 6996dee35698665ec42b5690.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70f21e209131cea4e442.md>)

Recorded evaluation configuration: {"Defense Method":"T2IShield\_CDA"}. This record does not assert independent reproduction.

### Personalization

Detection Accuracy: 50.7. Result ID: 6996dee35698665ec42b5684.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71911e209131cea5206b.md>)

Recorded evaluation configuration: {"Defense Method":"T2IShield\_CDA"}. This record does not assert independent reproduction.

### IBA

Detection Accuracy: 49.5. Result ID: 6996dee35698665ec42b5691.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a291e209131cea816e4.md>)

Recorded evaluation configuration: {"Defense Method":"NaviT2I"}. This record does not assert independent reproduction.

### IBA

Detection Accuracy: 49.5. Result ID: 6996dee35698665ec42b5694.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a291e209131cea816e4.md>)

Recorded evaluation configuration: {"Defense Method":"T2IShield\_CDA"}. This record does not assert independent reproduction.

### SemBD

Detection Accuracy: 48.5. Result ID: 6996dee35698665ec42b5698.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

Recorded evaluation configuration: {"Defense Method":"T2IShield\_CDA"}. This record does not assert independent reproduction.

### Rickrolling

Detection Accuracy: 47. Result ID: 6996dee35698665ec42b5687.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70471e209131cea4a212.md>)

Recorded evaluation configuration: {"Defense Method":"T2IShield\_FFT"}. This record does not assert independent reproduction.

### Personalization

Detection Accuracy: 45.8. Result ID: 6996dee35698665ec42b5683.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71911e209131cea5206b.md>)

Recorded evaluation configuration: {"Defense Method":"T2IShield\_FFT"}. This record does not assert independent reproduction.

### EvilEdit

Detection Accuracy: 45.5. Result ID: 6996dee35698665ec42b568a.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

Recorded evaluation configuration: {"Defense Method":"UFID"}. This record does not assert independent reproduction.

### Personalization

Detection Accuracy: 43. Result ID: 6996dee35698665ec42b5682.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71911e209131cea5206b.md>)

Recorded evaluation configuration: {"Defense Method":"UFID"}. This record does not assert independent reproduction.

### IBA

Detection Accuracy: 42.5. Result ID: 6996dee35698665ec42b5692.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a291e209131cea816e4.md>)

Recorded evaluation configuration: {"Defense Method":"UFID"}. This record does not assert independent reproduction.

### IBA

Detection Accuracy: 41. Result ID: 6996dee35698665ec42b5693.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a291e209131cea816e4.md>)

Recorded evaluation configuration: {"Defense Method":"T2IShield\_FFT"}. This record does not assert independent reproduction.

### SemBD

Detection Accuracy: 39.5. Result ID: 6996dee35698665ec42b5696.

[Reporting paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/698575a0c97a0b3e3b5d5cab.md>)

Recorded evaluation configuration: {"Defense Method":"UFID"}. This record does not assert independent reproduction.

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