# Text-to-Image Retrieval on MS-COCO (Hard)

Source: [Text-to-Image Retrieval on MS-COCO (Hard)](<https://www.sota2.com/research/sota/text-to-image-retrieval-on-ms-coco-hard>)

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

Data updated: 2026-02-26T06:01:28.580Z

Evaluation of Text-to-Image Retrieval performance using the Hard subset or difficulty setting of the MS-COCO dataset, measured using R@K 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 | Text-to-Image Retrieval |
| Dataset (including any stated split) | MS-COCO (Hard) |
| Metric | Recall@1 |
| 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

- [Recall@1](<https://www.sota2.com/research/sota/text-to-image-retrieval-on-ms-coco-hard.md?metric=Recall%401>): Higher is better; unit: PERCENT.

- [Recall@5](<https://www.sota2.com/research/sota/text-to-image-retrieval-on-ms-coco-hard.md?metric=Recall%405>): Higher is better; unit: PERCENT.

- [Recall@10](<https://www.sota2.com/research/sota/text-to-image-retrieval-on-ms-coco-hard.md?metric=Recall%4010>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Recall@1 | Result date | Report paper |
| --- | --- | --- | --- |
| BEIT-3 | 43.5 | Not recorded | 6997ea8209c33900fcd765c0 |
| Visual Model Checking + BEIT | 39.1 | Not recorded | 6997ea8209c33900fcd765c0 |
| SigLIP | 30.4 | Not recorded | 6997ea8209c33900fcd765c0 |
| CLIP | 28.3 | Not recorded | 6997ea8209c33900fcd765c0 |
| ALIGN | 28.3 | Not recorded | 6997ea8209c33900fcd765c0 |
| Visual Model Checking + ALIGN | 26.1 | Not recorded | 6997ea8209c33900fcd765c0 |
| Visual Model Checking (base) | 19.6 | Not recorded | 6997ea8209c33900fcd765c0 |
| CLIP | 15.2 | Not recorded | 6997ea8209c33900fcd765c0 |
| Visual Model Checking + CLIP | 4.3 | Not recorded | 6997ea8209c33900fcd765c0 |

### BEIT-3

Recall@1: 43.5. Result ID: 699b0b785698665ec49e550c.

[Reporting paper](<https://www.sota2.com/research/paper/6997ea8209c33900fcd765c0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d541e209131cea39310.md>)

[Implementation](<https://aka.ms/beit-3>)

Recorded evaluation configuration: {"Config":"COCO-Finetuned","Zero-Shot":false}. This record does not assert independent reproduction.

### Visual Model Checking + BEIT

Recall@1: 39.1. Result ID: 699b0b785698665ec49e5510.

[Reporting paper](<https://www.sota2.com/research/paper/6997ea8209c33900fcd765c0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6997ea8209c33900fcd765c0.md>)

Recorded evaluation configuration: {"Zero-Shot":false}. This record does not assert independent reproduction.

### SigLIP

Recall@1: 30.4. Result ID: 699b0b785698665ec49e550b.

[Reporting paper](<https://www.sota2.com/research/paper/6997ea8209c33900fcd765c0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70ae1e209131cea4c7a4.md>)

Recorded evaluation configuration: {"Parameters":"400M","Zero-Shot":true}. This record does not assert independent reproduction.

### CLIP

Recall@1: 28.3. Result ID: 699b0b785698665ec49e550a.

[Reporting paper](<https://www.sota2.com/research/paper/6997ea8209c33900fcd765c0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69581e209131cea21ad0.md>)

[Implementation](<https://github.com/OpenAI/CLIP>)

Recorded evaluation configuration: {"Backbone":"ViT-H","QuickGELU":true,"Zero-Shot":true}. This record does not assert independent reproduction.

### ALIGN

Recall@1: 28.3. Result ID: 699b0b785698665ec49e550d.

[Reporting paper](<https://www.sota2.com/research/paper/6997ea8209c33900fcd765c0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a201e209131cea261e7.md>)

Recorded evaluation configuration: {"Zero-Shot":true}. This record does not assert independent reproduction.

### Visual Model Checking + ALIGN

Recall@1: 26.1. Result ID: 699b0b785698665ec49e5511.

[Reporting paper](<https://www.sota2.com/research/paper/6997ea8209c33900fcd765c0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6997ea8209c33900fcd765c0.md>)

Recorded evaluation configuration: {"Zero-Shot":true}. This record does not assert independent reproduction.

### Visual Model Checking (base)

Recall@1: 19.6. Result ID: 699b0b785698665ec49e550e.

[Reporting paper](<https://www.sota2.com/research/paper/6997ea8209c33900fcd765c0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6997ea8209c33900fcd765c0.md>)

Recorded evaluation configuration: {"Zero-Shot":true}. This record does not assert independent reproduction.

### CLIP

Recall@1: 15.2. Result ID: 699b0b785698665ec49e5509.

[Reporting paper](<https://www.sota2.com/research/paper/6997ea8209c33900fcd765c0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69581e209131cea21ad0.md>)

[Implementation](<https://github.com/OpenAI/CLIP>)

Recorded evaluation configuration: {"Backbone":"ViT-B","Zero-Shot":true}. This record does not assert independent reproduction.

### Visual Model Checking + CLIP

Recall@1: 4.3. Result ID: 699b0b785698665ec49e550f.

[Reporting paper](<https://www.sota2.com/research/paper/6997ea8209c33900fcd765c0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6997ea8209c33900fcd765c0.md>)

Recorded evaluation configuration: {"Zero-Shot":true}. This record does not assert independent reproduction.

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