# Failure Prediction on ImageNet-LT (test)

Source: [Failure Prediction on ImageNet-LT (test)](<https://www.sota2.com/research/sota/failure-prediction-on-imagenet-lt-test>)

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

Data updated: 2026-02-26T05:25:20.807Z

Evaluation of model failure prediction performance on the test split of the ImageNet-Long Tail (LT) dataset. Metrics focus on calibration and discrimination capabilities, specifically AUC and ECE, segmented across head, medium, and tail classes.

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 | Failure Prediction |
| Dataset (including any stated split) | ImageNet-LT (test) |
| Metric | AUC (All) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 9 of 9 |
| 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

- [AUC (All)](<https://www.sota2.com/research/sota/failure-prediction-on-imagenet-lt-test.md?metric=AUC%20(All)>): Higher is better; unit: PERCENT.

- [AUC (Head)](<https://www.sota2.com/research/sota/failure-prediction-on-imagenet-lt-test.md?metric=AUC%20(Head)>): Higher is better; unit: PERCENT.

- [AUC (Med)](<https://www.sota2.com/research/sota/failure-prediction-on-imagenet-lt-test.md?metric=AUC%20(Med)>): Higher is better; unit: PERCENT.

- [AUC (Tail)](<https://www.sota2.com/research/sota/failure-prediction-on-imagenet-lt-test.md?metric=AUC%20(Tail)>): Higher is better; unit: PERCENT.

- [FPR-95 (All)](<https://www.sota2.com/research/sota/failure-prediction-on-imagenet-lt-test.md?metric=FPR-95%20(All)>): Lower is better; unit: PERCENT.

- [FPR-95 (Head)](<https://www.sota2.com/research/sota/failure-prediction-on-imagenet-lt-test.md?metric=FPR-95%20(Head)>): Lower is better; unit: PERCENT.

- [FPR-95 (Med)](<https://www.sota2.com/research/sota/failure-prediction-on-imagenet-lt-test.md?metric=FPR-95%20(Med)>): Lower is better; unit: PERCENT.

- [FPR-95 (Tail)](<https://www.sota2.com/research/sota/failure-prediction-on-imagenet-lt-test.md?metric=FPR-95%20(Tail)>): Lower is better; unit: PERCENT.

- [ECE (All)](<https://www.sota2.com/research/sota/failure-prediction-on-imagenet-lt-test.md?metric=ECE%20(All)>): Lower is better; unit: PERCENT.

- [ECE (Head)](<https://www.sota2.com/research/sota/failure-prediction-on-imagenet-lt-test.md?metric=ECE%20(Head)>): Lower is better; unit: PERCENT.

- [ECE (Med)](<https://www.sota2.com/research/sota/failure-prediction-on-imagenet-lt-test.md?metric=ECE%20(Med)>): Lower is better; unit: PERCENT.

- [ECE (Tail)](<https://www.sota2.com/research/sota/failure-prediction-on-imagenet-lt-test.md?metric=ECE%20(Tail)>): Lower is better; unit: PERCENT.

## Recorded results

| Method | AUC (All) | Result date | Report paper |
| --- | --- | --- | --- |
| TLC | 67.2 | Not recorded | 68fa6c191e209131cea322d9 |
| LDAM-DRW | 66.8 | Not recorded | 68fa6c191e209131cea322d9 |
| TLC | 66.7 | Not recorded | 68fa6c191e209131cea322d9 |
| TLC | 66.7 | Not recorded | 68fa6c191e209131cea322d9 |
| cRT | 66.4 | Not recorded | 68fa6c191e209131cea322d9 |
| RIDE | 66.2 | Not recorded | 68fa6c191e209131cea322d9 |
| tau-norm | 66.1 | Not recorded | 68fa6c191e209131cea322d9 |
| OLTR | 66 | Not recorded | 68fa6c191e209131cea322d9 |
| Focal Loss | 65.4 | Not recorded | 68fa6c191e209131cea322d9 |

### TLC

AUC (All): 67.2. Result ID: 698ed352ffc3901de0a1c39a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c191e209131cea322d9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c191e209131cea322d9.md>)

[Implementation](<https://github.com/lblaoke/TLC>)

Recorded evaluation configuration: {"Number of experts":4}. This record does not assert independent reproduction.

### LDAM-DRW

AUC (All): 66.8. Result ID: 698ed352ffc3901de0a1c394.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c191e209131cea322d9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66901e209131cea10e0f.md>)

[Implementation](<https://github.com/kaidic/LDAM-DRW>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### TLC

AUC (All): 66.7. Result ID: 698ed352ffc3901de0a1c398.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c191e209131cea322d9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c191e209131cea322d9.md>)

[Implementation](<https://github.com/lblaoke/TLC>)

Recorded evaluation configuration: {"Number of experts":2}. This record does not assert independent reproduction.

### TLC

AUC (All): 66.7. Result ID: 698ed352ffc3901de0a1c399.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c191e209131cea322d9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c191e209131cea322d9.md>)

[Implementation](<https://github.com/lblaoke/TLC>)

Recorded evaluation configuration: {"Number of experts":3}. This record does not assert independent reproduction.

### cRT

AUC (All): 66.4. Result ID: 698ed352ffc3901de0a1c396.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c191e209131cea322d9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa671b1e209131cea144a9.md>)

[Implementation](<https://github.com/facebookresearch/classifier-balancing>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### RIDE

AUC (All): 66.2. Result ID: 698ed352ffc3901de0a1c397.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c191e209131cea322d9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c7b1e209131cea3436b.md>)

[Implementation](<https://github.com/frank-xwang/RIDE-LongTailRecognition>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### tau-norm

AUC (All): 66.1. Result ID: 698ed352ffc3901de0a1c395.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c191e209131cea322d9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa671b1e209131cea144a9.md>)

[Implementation](<https://github.com/facebookresearch/classifier-balancing>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### OLTR

AUC (All): 66. Result ID: 698ed352ffc3901de0a1c393.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c191e209131cea322d9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa659f1e209131cea0b4aa.md>)

[Implementation](<https://liuziwei7.github.io/projects/LongTail.html>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Focal Loss

AUC (All): 65.4. Result ID: 698ed352ffc3901de0a1c392.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6c191e209131cea322d9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa640c1e209131cea02556.md>)

[Implementation](<https://github.com/facebookresearch/Detectron>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

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