# Image Retrieval on MS COCO Karpathy's split (val)

Source: [Image Retrieval on MS COCO Karpathy's split (val)](<https://www.sota2.com/research/sota/image-retrieval-on-ms-coco-karpathy-s-split-val>)

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

Data updated: 2026-05-08T08:08:26.759Z

Evaluation of Image Retrieval performance using Karpathy's specific validation split of the MS COCO dataset, measured using Deletion, Insertion, and IMD 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 | Image Retrieval |
| Dataset (including any stated split) | MS COCO Karpathy's split (val) |
| Metric | Deletion @1 |
| Unit | PERCENT |
| Direction | Lower is better |
| Results loaded | 15 of 15 |
| 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

- [Deletion @1](<https://www.sota2.com/research/sota/image-retrieval-on-ms-coco-karpathy-s-split-val.md?metric=Deletion%20%401>): Lower is better; unit: PERCENT.

- [Deletion @5](<https://www.sota2.com/research/sota/image-retrieval-on-ms-coco-karpathy-s-split-val.md?metric=Deletion%20%405>): Lower is better; unit: PERCENT.

- [Insertion @1](<https://www.sota2.com/research/sota/image-retrieval-on-ms-coco-karpathy-s-split-val.md?metric=Insertion%20%401>): Lower is better; unit: PERCENT.

- [Insertion @5](<https://www.sota2.com/research/sota/image-retrieval-on-ms-coco-karpathy-s-split-val.md?metric=Insertion%20%405>): Lower is better; unit: PERCENT.

- [IMD @1](<https://www.sota2.com/research/sota/image-retrieval-on-ms-coco-karpathy-s-split-val.md?metric=IMD%20%401>): Higher is better; unit: SCALAR.

- [IMD @5](<https://www.sota2.com/research/sota/image-retrieval-on-ms-coco-karpathy-s-split-val.md?metric=IMD%20%405>): Higher is better; unit: SCALAR.

- [Deletion](<https://www.sota2.com/research/sota/image-retrieval-on-ms-coco-karpathy-s-split-val.md?metric=Deletion>): Lower is better; unit: PERCENT.

- [Insertion](<https://www.sota2.com/research/sota/image-retrieval-on-ms-coco-karpathy-s-split-val.md?metric=Insertion>): Lower is better; unit: PERCENT.

- [IMD](<https://www.sota2.com/research/sota/image-retrieval-on-ms-coco-karpathy-s-split-val.md?metric=IMD>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Deletion @1 | Result date | Report paper |
| --- | --- | --- | --- |
| Grad-ECLIP | 12.5 | Not recorded | 68fa772a1e209131cea71801 |
| MaskCLIP | 13.2 | Not recorded | 68fa772a1e209131cea71801 |
| Grad-ECLIP w/o λi | 13.9 | Not recorded | 68fa772a1e209131cea71801 |
| raw attention | 17.1 | Not recorded | 68fa772a1e209131cea71801 |
| GAME | 17.1 | Not recorded | 68fa772a1e209131cea71801 |
| Grad-CAM | 17.2 | Not recorded | 68fa772a1e209131cea71801 |
| CLIPSurgery | 17.9 | Not recorded | 68fa772a1e209131cea71801 |
| M2IB | 17.9 | Not recorded | 68fa772a1e209131cea71801 |
| Rollout | 19.5 | Not recorded | 68fa772a1e209131cea71801 |

### Grad-ECLIP

Deletion @1: 12.5. Result ID: 69fd99f047f34ce924e07793.

[Reporting paper](<https://www.sota2.com/research/paper/68fa772a1e209131cea71801.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa772a1e209131cea71801.md>)

Recorded evaluation configuration: {"Backbone":"CLIP"}. This record does not assert independent reproduction.

### MaskCLIP

Deletion @1: 13.2. Result ID: 69fd99f047f34ce924e07790.

[Reporting paper](<https://www.sota2.com/research/paper/68fa772a1e209131cea71801.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d1f1e209131cea37ece.md>)

Recorded evaluation configuration: {"Backbone":"CLIP"}. This record does not assert independent reproduction.

### Grad-ECLIP w/o λi

Deletion @1: 13.9. Result ID: 69fd99f047f34ce924e07794.

[Reporting paper](<https://www.sota2.com/research/paper/68fa772a1e209131cea71801.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa772a1e209131cea71801.md>)

Recorded evaluation configuration: {"Backbone":"CLIP"}. This record does not assert independent reproduction.

### raw attention

Deletion @1: 17.1. Result ID: 69fd99f047f34ce924e0778c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa772a1e209131cea71801.md>)

Recorded evaluation configuration: {"Backbone":"CLIP"}. This record does not assert independent reproduction.

### GAME

Deletion @1: 17.1. Result ID: 69fd99f047f34ce924e0778f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa772a1e209131cea71801.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69a11e209131cea233a8.md>)

Recorded evaluation configuration: {"Backbone":"CLIP"}. This record does not assert independent reproduction.

### Grad-CAM

Deletion @1: 17.2. Result ID: 69fd99f047f34ce924e0778e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa772a1e209131cea71801.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66cb1e209131cea12568.md>)

[Implementation](<https://github.com/ramprs/grad-cam/>)

Recorded evaluation configuration: {"Backbone":"CLIP"}. This record does not assert independent reproduction.

### CLIPSurgery

Deletion @1: 17.9. Result ID: 69fd99f047f34ce924e07791.

[Reporting paper](<https://www.sota2.com/research/paper/68fa772a1e209131cea71801.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74dd1e209131cea64460.md>)

Recorded evaluation configuration: {"Backbone":"CLIP"}. This record does not assert independent reproduction.

### M2IB

Deletion @1: 17.9. Result ID: 69fd99f047f34ce924e07792.

[Reporting paper](<https://www.sota2.com/research/paper/68fa772a1e209131cea71801.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73b61e209131cea5e107.md>)

[Implementation](<https://github.com/YingWANGG/M2IB>)

Recorded evaluation configuration: {"Backbone":"CLIP"}. This record does not assert independent reproduction.

### Rollout

Deletion @1: 19.5. Result ID: 69fd99f047f34ce924e0778d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa772a1e209131cea71801.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa67ae1e209131cea17be6.md>)

Recorded evaluation configuration: {"Backbone":"CLIP"}. This record does not assert independent reproduction.

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