# Visual Question Answering on NaturalBench

Source: [Visual Question Answering on NaturalBench](<https://www.sota2.com/research/sota/visual-question-answering-on-naturalbench>)

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

Data updated: 2026-02-26T06:04:22.740Z

Evaluation of model performance on the NaturalBench dataset for Visual Question Answering, reporting Simple accuracy, Paired accuracy, and Yes-rate.

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 | Visual Question Answering |
| Dataset (including any stated split) | NaturalBench |
| Metric | Simple Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 2 of 2 |
| 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

- [Simple Accuracy](<https://www.sota2.com/research/sota/visual-question-answering-on-naturalbench.md?metric=Simple%20Accuracy>): Higher is better; unit: PERCENT.

- [Paired Accuracy](<https://www.sota2.com/research/sota/visual-question-answering-on-naturalbench.md?metric=Paired%20Accuracy>): Higher is better; unit: PERCENT.

- [Yes-Rate](<https://www.sota2.com/research/sota/visual-question-answering-on-naturalbench.md?metric=Yes-Rate>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Simple Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| LLaVA-1.5 13B | 0.6946 | Not recorded | 69705dade25ad2fa9d1ed310 |
| LLaVA-1.5 13B | 0.6836 | Not recorded | 69705dade25ad2fa9d1ed310 |

### LLaVA-1.5 13B

Simple Accuracy: 0.6946. Result ID: 69967cfd5698665ec4c4d370.

[Reporting paper](<https://www.sota2.com/research/paper/69705dade25ad2fa9d1ed310.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69705dade25ad2fa9d1ed310.md>)

Recorded evaluation configuration: {"Intervention Strategy":"No-intervention baseline"}. This record does not assert independent reproduction.

### LLaVA-1.5 13B

Simple Accuracy: 0.6836. Result ID: 69967cfd5698665ec4c4d371.

[Reporting paper](<https://www.sota2.com/research/paper/69705dade25ad2fa9d1ed310.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69705dade25ad2fa9d1ed310.md>)

Recorded evaluation configuration: {"Intervention Strategy":"Q4 system redistribution (prop)"}. This record does not assert independent reproduction.

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