# Open-Set Recognition on ImageNet-hard

Source: [Open-Set Recognition on ImageNet-hard](<https://www.sota2.com/research/sota/open-set-recognition-on-imagenet-hard>)

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

Data updated: 2026-05-19T11:42:37.212Z

Evaluation of Open-Set Recognition performance on the ImageNet-hard dataset configuration, reporting KRR and HC-FKAR metrics.

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 | Open-Set Recognition |
| Dataset (including any stated split) | ImageNet-hard |
| Metric | KRR |
| Unit | SCALAR |
| 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

- [KRR](<https://www.sota2.com/research/sota/open-set-recognition-on-imagenet-hard.md?metric=KRR>): Higher is better; unit: SCALAR.

- [HC-FKAR](<https://www.sota2.com/research/sota/open-set-recognition-on-imagenet-hard.md?metric=HC-FKAR>): Higher is better; unit: SCALAR.

- [Known Accuracy](<https://www.sota2.com/research/sota/open-set-recognition-on-imagenet-hard.md?metric=Known%20Accuracy>): Higher is better; unit: PERCENT.

- [FKAR](<https://www.sota2.com/research/sota/open-set-recognition-on-imagenet-hard.md?metric=FKAR>): Higher is better; unit: SCALAR.

## Recorded results

| Method | KRR | Result date | Report paper |
| --- | --- | --- | --- |
| MSP | 0.263 | Not recorded | 6a0beef0dbc836b87d82c8ec |
| EGUR-A | 0.238 | Not recorded | 6a0beef0dbc836b87d82c8ec |
| Energy | 0.164 | Not recorded | 6a0beef0dbc836b87d82c8ec |
| NNGuide | 0.139 | Not recorded | 6a0beef0dbc836b87d82c8ec |
| ViM | 0.13 | Not recorded | 6a0beef0dbc836b87d82c8ec |
| CADRef | 0.1 | Not recorded | 6a0beef0dbc836b87d82c8ec |
| NCI-OOD | 0.055 | Not recorded | 6a0beef0dbc836b87d82c8ec |
| Prototype distance | 0.053 | Not recorded | 6a0beef0dbc836b87d82c8ec |
| kNN distance | 0.05 | Not recorded | 6a0beef0dbc836b87d82c8ec |

### MSP

KRR: 0.263. Result ID: 6a0c4c01fbc373592c036e90.

[Reporting paper](<https://www.sota2.com/research/paper/6a0beef0dbc836b87d82c8ec.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa64e41e209131cea0731a.md>)

[Implementation](<https://www.sota2.com/github.com/hendrycks/error-detection>)

Recorded evaluation configuration: {"Threshold":"natural/default"}. This record does not assert independent reproduction.

### EGUR-A

KRR: 0.238. Result ID: 6a0c4c01fbc373592c036e96.

[Reporting paper](<https://www.sota2.com/research/paper/6a0beef0dbc836b87d82c8ec.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0beef0dbc836b87d82c8ec.md>)

Recorded evaluation configuration: {"Threshold":"calibrated operating point"}. This record does not assert independent reproduction.

### Energy

KRR: 0.164. Result ID: 6a0c4c01fbc373592c036e91.

[Reporting paper](<https://www.sota2.com/research/paper/6a0beef0dbc836b87d82c8ec.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69cc1e209131cea24242.md>)

Recorded evaluation configuration: {"Threshold":"natural/default"}. This record does not assert independent reproduction.

### NNGuide

KRR: 0.139. Result ID: 6a0c4c01fbc373592c036e92.

[Reporting paper](<https://www.sota2.com/research/paper/6a0beef0dbc836b87d82c8ec.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70ad1e209131cea4c735.md>)

[Implementation](<https://github.com/roomo7time/nnguide>)

Recorded evaluation configuration: {"Threshold":"natural/default"}. This record does not assert independent reproduction.

### ViM

KRR: 0.13. Result ID: 6a0c4c01fbc373592c036e95.

[Reporting paper](<https://www.sota2.com/research/paper/6a0beef0dbc836b87d82c8ec.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c151e209131cea32091.md>)

[Implementation](<https://github.com/haoqiwang/vim>)

Recorded evaluation configuration: {"Threshold":"natural/default"}. This record does not assert independent reproduction.

### CADRef

KRR: 0.1. Result ID: 6a0c4c01fbc373592c036e94.

[Reporting paper](<https://www.sota2.com/research/paper/6a0beef0dbc836b87d82c8ec.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77761e209131cea734a1.md>)

Recorded evaluation configuration: {"Threshold":"natural/default"}. This record does not assert independent reproduction.

### NCI-OOD

KRR: 0.055. Result ID: 6a0c4c01fbc373592c036e93.

[Reporting paper](<https://www.sota2.com/research/paper/6a0beef0dbc836b87d82c8ec.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77c81e209131cea754d6.md>)

Recorded evaluation configuration: {"Threshold":"natural/default"}. This record does not assert independent reproduction.

### Prototype distance

KRR: 0.053. Result ID: 6a0c4c01fbc373592c036e9e.

[Reporting paper](<https://www.sota2.com/research/paper/6a0beef0dbc836b87d82c8ec.md>)

Recorded evaluation configuration: {"Workpoint":"default","Confidence Source":"prototype-derived candidate confidence"}. This record does not assert independent reproduction.

### kNN distance

KRR: 0.05. Result ID: 6a0c4c01fbc373592c036e9d.

[Reporting paper](<https://www.sota2.com/research/paper/6a0beef0dbc836b87d82c8ec.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e2f1e209131cea3e69b.md>)

[Implementation](<https://github.com/deeplearning-wisc/knn-ood>)

Recorded evaluation configuration: {"Workpoint":"default","Confidence Source":"prototype-derived candidate confidence"}. This record does not assert independent reproduction.

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