# Classification Attribution on ImageNet misclassified samples (val)

Source: [Classification Attribution on ImageNet misclassified samples (val)](<https://www.sota2.com/research/sota/classification-attribution-on-imagenet-misclassified-samples-val>)

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

Data updated: 2026-06-17T07:52:55.006Z

Evaluation of classification attribution methods applied specifically to samples previously misclassified within the ImageNet validation set, measured using fidelity metrics such as Insertion (Ins.), Deletion (Del.), MEC, and A-C.

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 | Classification Attribution |
| Dataset (including any stated split) | ImageNet misclassified samples (val) |
| Metric | Insertion Score |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 7 of 7 |
| 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

- [Insertion Score](<https://www.sota2.com/research/sota/classification-attribution-on-imagenet-misclassified-samples-val.md?metric=Insertion%20Score>): Higher is better; unit: SCALAR.

- [Deletion Score](<https://www.sota2.com/research/sota/classification-attribution-on-imagenet-misclassified-samples-val.md?metric=Deletion%20Score>): Higher is better; unit: SCALAR.

- [Average High Score](<https://www.sota2.com/research/sota/classification-attribution-on-imagenet-misclassified-samples-val.md?metric=Average%20High%20Score>): Higher is better; unit: SCALAR.

- [Score @ 50% Threshold](<https://www.sota2.com/research/sota/classification-attribution-on-imagenet-misclassified-samples-val.md?metric=Score%20%40%2050%25%20Threshold>): Higher is better; unit: SCALAR.

- [MEC](<https://www.sota2.com/research/sota/classification-attribution-on-imagenet-misclassified-samples-val.md?metric=MEC>): Lower is better; unit: SCALAR.

- [A-C Score](<https://www.sota2.com/research/sota/classification-attribution-on-imagenet-misclassified-samples-val.md?metric=A-C%20Score>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Insertion Score | Result date | Report paper |
| --- | --- | --- | --- |
| Greedy | 0.4831 | Not recorded | 6a322a57d33dfb23e783ed5d |
| PhaseWin | 0.4264 | Not recorded | 6a322a57d33dfb23e783ed5d |
| RISE | 0.2543 | Not recorded | 6a322a57d33dfb23e783ed5d |
| HSIC | 0.2365 | Not recorded | 6a322a57d33dfb23e783ed5d |
| Gradient Integral | 0.193 | Not recorded | 6a322a57d33dfb23e783ed5d |
| Gradient | 0.1891 | Not recorded | 6a322a57d33dfb23e783ed5d |
| IGOS++ | 0.153 | Not recorded | 6a322a57d33dfb23e783ed5d |

### Greedy

Insertion Score: 0.4831. Result ID: 6a325254fbc373592c2fa646.

[Reporting paper](<https://www.sota2.com/research/paper/6a322a57d33dfb23e783ed5d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74c31e209131cea63a16.md>)

[Implementation](<https://github.com/RuoyuChen10/SMDL-Attribution>)

Recorded evaluation configuration: {"Backbone":"CLIP RN101","Attribution Target":"Model's wrong prediction"}. This record does not assert independent reproduction.

### PhaseWin

Insertion Score: 0.4264. Result ID: 6a325254fbc373592c2fa647.

[Reporting paper](<https://www.sota2.com/research/paper/6a322a57d33dfb23e783ed5d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a322a57d33dfb23e783ed5d.md>)

Recorded evaluation configuration: {"Backbone":"CLIP RN101","Attribution Target":"Model's wrong prediction"}. This record does not assert independent reproduction.

### RISE

Insertion Score: 0.2543. Result ID: 6a325254fbc373592c2fa644.

[Reporting paper](<https://www.sota2.com/research/paper/6a322a57d33dfb23e783ed5d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa64dc1e209131cea06cb9.md>)

[Implementation](<https://github.com/pytorch/vision>)

Recorded evaluation configuration: {"Backbone":"CLIP RN101","Attribution Target":"Model's wrong prediction"}. This record does not assert independent reproduction.

### HSIC

Insertion Score: 0.2365. Result ID: 6a325254fbc373592c2fa645.

[Reporting paper](<https://www.sota2.com/research/paper/6a322a57d33dfb23e783ed5d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d891e209131cea3a77b.md>)

[Implementation](<https://github.com/paulnovello/HSIC-Attribution-Method>)

Recorded evaluation configuration: {"Backbone":"CLIP RN101","Attribution Target":"Model's wrong prediction"}. This record does not assert independent reproduction.

### Gradient Integral

Insertion Score: 0.193. Result ID: 6a325254fbc373592c2fa642.

[Reporting paper](<https://www.sota2.com/research/paper/6a322a57d33dfb23e783ed5d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63661e209131ce9fe6bd.md>)

Recorded evaluation configuration: {"Backbone":"CLIP RN101","Attribution Target":"Model's wrong prediction"}. This record does not assert independent reproduction.

### Gradient

Insertion Score: 0.1891. Result ID: 6a325254fbc373592c2fa641.

[Reporting paper](<https://www.sota2.com/research/paper/6a322a57d33dfb23e783ed5d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa614c1e209131ce9f38a9.md>)

Recorded evaluation configuration: {"Backbone":"CLIP RN101","Attribution Target":"Model's wrong prediction"}. This record does not assert independent reproduction.

### IGOS++

Insertion Score: 0.153. Result ID: 6a325254fbc373592c2fa643.

[Reporting paper](<https://www.sota2.com/research/paper/6a322a57d33dfb23e783ed5d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69d21e209131cea2462c.md>)

Recorded evaluation configuration: {"Backbone":"CLIP RN101","Attribution Target":"Model's wrong prediction"}. This record does not assert independent reproduction.

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