---
license: apache-2.0
task_categories:
- visual-question-answering
- question-answering
language:
- en
tags:
- neuroscience
- fmri
- vqa
- brain-decoding
pretty_name: NSD-VQA
size_categories:
- 1M<n<10M
---

# NSD-VQA

**NSD-VQA** is a large-scale visual question answering benchmark for studying what visual and semantic information can be decoded from human fMRI responses to natural images.

It is introduced in **Brain-IT-VQA: From Brain Signals to Answers**.

🔗 **Project page:** https://mcosarinsky.github.io/brain-it-vqa/

## Overview

NSD-VQA is built from the **Natural Scenes Dataset (NSD)** and provides automatically generated question-answer annotations grounded in NSD images.

The dataset contains approximately:

- **73K NSD images**
- **~20 question-answer pairs per image**
- **20 controlled semantic question categories**

The benchmark is designed for fine-grained evaluation of visual and semantic decoding from brain activity, beyond a single aggregate VQA score.

Question categories include:

- object recognition
- counting
- color
- actions
- spatial position
- scene understanding
- human-object interactions
- semantic categories such as animals, vehicles, food, appliances, and household objects

This repository contains **only generated annotations and metadata**.  
It does **not** redistribute NSD images or fMRI recordings.

Users must obtain the original NSD dataset separately under its original terms.

## Dataset Variants

This repository includes two variants:

| File | Description |
|---|---|
| `nsd_vqa.parquet` | Short-answer NSD-VQA annotations |
| `nsd_vqa_fs.parquet` | Full-sentence answer variant, NSD-VQA-FS |

Both files use the same schema and contain one row per question-answer pair.


## Dataset Structure

Each parquet file contains the following columns:

| Column | Description |
|---|---|
| `nsd_image_id` | NSD image identifier |
| `coco_image_id` | Corresponding COCO image identifier |
| `question_id` | Unique question identifier |
| `answer_id` | Unique answer identifier |
| `question` | Natural language question |
| `answers` | Ground-truth answer |
| `category` | Controlled semantic question category |

---

# Citation

If you use this dataset, please cite:

```bibtex
@article{beliy2026brainitvqa,
  title={Brain-IT-VQA: From Brain Signals to Answers},
  author={Beliy, Roman and others},
  year={2026}
}
```