# Visual Question Answering on SV-QA HRBench-4K

Source: [Visual Question Answering on SV-QA HRBench-4K](<https://www.sota2.com/research/sota/visual-question-answering-on-sv-qa-hrbench-4k>)

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

Data updated: 2026-05-20T08:03:50.709Z

Evaluation of model performance on the SV-QA HRBench-4K dataset for Visual Question Answering, reporting Q1 accuracy, Q2 accuracy, and Both Correct 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 | Visual Question Answering |
| Dataset (including any stated split) | SV-QA HRBench-4K |
| Metric | Q1 Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 6 of 6 |
| 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

- [Q1 Accuracy](<https://www.sota2.com/research/sota/visual-question-answering-on-sv-qa-hrbench-4k.md?metric=Q1%20Accuracy>): Higher is better; unit: PERCENT.

- [Q2 Accuracy](<https://www.sota2.com/research/sota/visual-question-answering-on-sv-qa-hrbench-4k.md?metric=Q2%20Accuracy>): Higher is better; unit: PERCENT.

- [Both Correct](<https://www.sota2.com/research/sota/visual-question-answering-on-sv-qa-hrbench-4k.md?metric=Both%20Correct>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Q1 Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| DeepEyes | 71.3 | Not recorded | 6a0d4002f984739ed6737774 |
| SLVR-7B | 70.4 | Not recorded | 6a0d4002f984739ed6737774 |
| LVR | 70.1 | Not recorded | 6a0d4002f984739ed6737774 |
| Visual-SR1 | 63.9 | Not recorded | 6a0d4002f984739ed6737774 |
| Qwen2.5-VL-7B | 60.5 | Not recorded | 6a0d4002f984739ed6737774 |
| GRIT | 58.4 | Not recorded | 6a0d4002f984739ed6737774 |

### DeepEyes

Q1 Accuracy: 71.3. Result ID: 6a0d6a64fbc373592c054c2f.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4002f984739ed6737774.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78c71e209131cea7a75e.md>)

Recorded evaluation configuration: {"Reasoning Paradigm":"Text + Visual Cropping"}. This record does not assert independent reproduction.

### SLVR-7B

Q1 Accuracy: 70.4. Result ID: 6a0d6a64fbc373592c054c32.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4002f984739ed6737774.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0d4002f984739ed6737774.md>)

Recorded evaluation configuration: {"Reasoning Paradigm":"Visual + Semantic Latent"}. This record does not assert independent reproduction.

### LVR

Q1 Accuracy: 70.1. Result ID: 6a0d6a64fbc373592c054c31.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4002f984739ed6737774.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7be71e209131cea8b35b.md>)

Recorded evaluation configuration: {"Reasoning Paradigm":"Visual Latent"}. This record does not assert independent reproduction.

### Visual-SR1

Q1 Accuracy: 63.9. Result ID: 6a0d6a64fbc373592c054c30.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4002f984739ed6737774.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7ae11e209131cea85a1c.md>)

Recorded evaluation configuration: {"Reasoning Paradigm":"Text CoT"}. This record does not assert independent reproduction.

### Qwen2.5-VL-7B

Q1 Accuracy: 60.5. Result ID: 6a0d6a64fbc373592c054c2d.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4002f984739ed6737774.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d3c.md>)

Recorded evaluation configuration: {"Reasoning Paradigm":"Text-only"}. This record does not assert independent reproduction.

### GRIT

Q1 Accuracy: 58.4. Result ID: 6a0d6a64fbc373592c054c2e.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4002f984739ed6737774.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78bd1e209131cea7a2a3.md>)

Recorded evaluation configuration: {"Reasoning Paradigm":"Text CoT"}. This record does not assert independent reproduction.

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