# VQA on GQA (held-out)

Source: [VQA on GQA (held-out)](<https://www.sota2.com/research/sota/vqa-on-gqa-held-out>)

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

Data updated: 2026-02-26T05:29:34.605Z

Evaluation of model performance on the held-out split of the GQA dataset for Visual Question Answering tasks, reporting Accuracy.

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 | VQA |
| Dataset (including any stated split) | GQA (held-out) |
| Metric | Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 3 of 3 |
| 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

- [Accuracy](<https://www.sota2.com/research/sota/vqa-on-gqa-held-out.md?metric=Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| VisReP | 59.2 | Not recorded | 68fa72cc1e209131cea5884e |
| Frozen Proprietary LLM | 53.9 | Not recorded | 68fa72cc1e209131cea5884e |
| Frozen Open LLM | 50 | Not recorded | 68fa72cc1e209131cea5884e |

### VisReP

Accuracy: 59.2. Result ID: 69927b623a6c3148490f424b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72cc1e209131cea5884e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72cc1e209131cea5884e.md>)

[Implementation](<https://zaidkhan.me/ViReP>)

Recorded evaluation configuration: {"LLM":"CodeLlama-7B","Backbone":"ViperGPT \[29\]"}. This record does not assert independent reproduction.

### Frozen Proprietary LLM

Accuracy: 53.9. Result ID: 69927b623a6c3148490f4249.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72cc1e209131cea5884e.md>)

Recorded evaluation configuration: {"LLM":"GPT-3.5-turbo","Backbone":"ViperGPT \[29\]"}. This record does not assert independent reproduction.

### Frozen Open LLM

Accuracy: 50. Result ID: 69927b623a6c3148490f424a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72cc1e209131cea5884e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71ff1e209131cea53ff7.md>)

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

Recorded evaluation configuration: {"LLM":"CodeLlama-7B","Backbone":"ViperGPT \[29\]"}. This record does not assert independent reproduction.

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