# Visual Question Answering on VQA v2 (test-dev) (Robustness)

Source: [Visual Question Answering on VQA v2 (test-dev) (Robustness)](<https://www.sota2.com/research/sota/visual-question-answering-on-vqa-v2-test-dev-robustness>)

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

Data updated: 2026-04-24T08:03:31.157Z

Evaluation of model performance on the VQA v2 test-dev split focusing on robustness metrics across various input corruptions (Noise, Blur, Weather, Digital).

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) | VQA v2 (test-dev) |
| Metric | Accuracy (Noise) |
| 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

- [Accuracy (Noise)](<https://www.sota2.com/research/sota/visual-question-answering-on-vqa-v2-test-dev-robustness.md?metric=Accuracy%20(Noise)>): Higher is better; unit: PERCENT.

- [Accuracy (Blur)](<https://www.sota2.com/research/sota/visual-question-answering-on-vqa-v2-test-dev-robustness.md?metric=Accuracy%20(Blur)>): Higher is better; unit: PERCENT.

- [Accuracy (Weather)](<https://www.sota2.com/research/sota/visual-question-answering-on-vqa-v2-test-dev-robustness.md?metric=Accuracy%20(Weather)>): Higher is better; unit: PERCENT.

- [Accuracy (Digital)](<https://www.sota2.com/research/sota/visual-question-answering-on-vqa-v2-test-dev-robustness.md?metric=Accuracy%20(Digital)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy (Noise) | Result date | Report paper |
| --- | --- | --- | --- |
| Latent Denoising | 71.1 | Not recorded | 69eaf92a55a3c32af89f89e0 |
| Baseline | 69.4 | Not recorded | 69eaf92a55a3c32af89f89e0 |

### Latent Denoising

Accuracy (Noise): 71.1. Result ID: 69eb23c447f34ce924d32ff3.

[Reporting paper](<https://www.sota2.com/research/paper/69eaf92a55a3c32af89f89e0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69eaf92a55a3c32af89f89e0.md>)

Recorded evaluation configuration: {"Backbone":"LLaVA+CLIP","Corruption Severity":"5"}. This record does not assert independent reproduction.

### Baseline

Accuracy (Noise): 69.4. Result ID: 69eb23c447f34ce924d32ff2.

[Reporting paper](<https://www.sota2.com/research/paper/69eaf92a55a3c32af89f89e0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa717d1e209131cea51970.md>)

[Implementation](<https://llava-vl.github.io/>)

Recorded evaluation configuration: {"Backbone":"LLaVA+CLIP","Corruption Severity":"5"}. This record does not assert independent reproduction.

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