---
license: mit
language:
- en
---


<h1 align="center">Vis-IR: Unifying Search With Visualized Information Retrieval</h1>

<p align="center">
    <a href="https://arxiv.org/abs/2502.11431">
        <img alt="Build" src="http://img.shields.io/badge/arXiv-2502.11431-B31B1B.svg">
    </a>
    <a href="https://github.com/VectorSpaceLab/Vis-IR">
        <img alt="Build" src="https://img.shields.io/badge/Github-Code-blue">
    </a>
    <a href="https://huggingface.co/datasets/marsh123/VIRA/">
        <img alt="Build" src="https://img.shields.io/badge/🤗 Datasets-VIRA-yellow">
    </a>  
    <a href="https://huggingface.co/datasets/marsh123/MVRB">
        <img alt="Build" src="https://img.shields.io/badge/🤗 Datasets-MVRB-yellow">
    </a>  
    <!-- <a href="">
        <img alt="Build" src="https://img.shields.io/badge/🤗 Model-UniSE CLIP-yellow">
    </a>  -->
    <a href="https://huggingface.co/marsh123/UniSE">
        <img alt="Build" src="https://img.shields.io/badge/🤗 Model-UniSE MLLM-yellow">
    </a> 
     
</p>

## Overview

**VIRA** (Vis-IR Aggregation), a large-scale dataset comprising a vast collection of screenshots from diverse sources, carefully curated into captioned and questionanswer formats.

## Statistics

There are three types of data in VIRA: caption data, query-to-screenshot (q2s) data, and screenshot+query-to-screenshot (sq2s) data. The table below provides a detailed breakdown of the data counts for each domain and type.

![image/png](https://cdn-uploads.huggingface.co/production/uploads/66164f6245336ca774679611/EXXiP6zykuQunrx30hwBt.png)


## Organization Structure

The dataset is organized in the following structure:

```tree
Domain/  
├── caption.jsonl: a screenshot image path and its corresponding caption 
├── q2s.jsonl: a query, a positive screenshot and eight negative screenshots
├── sq2s.jsonl: a query, a query screenshot, a positive screenshot and eight negative screenshots
├── Domain.tar.gz.partaa 
├── Domain.tar.gz.partab
├── Domain.tar.gz.partac
... 
```

## Download Images
We have released all the images. Due to the large total size, they are split into multiple parts. You can download the images using the script below:

```bash
#!/bin/bash

FOLDER="Chart" # “News” "PDFA" "Papers" "Product" "Readmes" "Wiki"
BASE_URL="https://huggingface.co/datasets/marsh123/VIRA/resolve/main/${FOLDER}"

mkdir -p "$FOLDER"
cd "$FOLDER"

for first in {a..z}; do
  for second in {a..z}; do
    part="${first}${second}"
    file="${FOLDER}.tar.gz.part${part}"
    url="${BASE_URL}/${file}"

    echo "Trying: $url"
    if curl --output /dev/null --silent --head --fail "$url"; then
      wget --continue "$url"
    else
      echo "No more parts after part${part}"
      break 2  
    fi
  done
done

echo "Merging parts..."
cat "${FOLDER}.tar.gz.part"* > "${FOLDER}.tar.gz"

echo "Extracting..."
tar -xzvf "${FOLDER}.tar.gz"

echo "Cleaning up..."
rm "${FOLDER}.tar.gz.part"*
rm "${FOLDER}.tar.gz"

echo "✅ Done!"

```


## License
VIRA is licensed under the [MIT License](LICENSE). 


## Citation
If you find this dataset useful, please cite:

```
@article{liu2025any,
  title={Any Information Is Just Worth One Single Screenshot: Unifying Search With Visualized Information Retrieval},
  author={Liu, Ze and Liang, Zhengyang and Zhou, Junjie and Liu, Zheng and Lian, Defu},
  journal={arXiv preprint arXiv:2502.11431},
  year={2025}
}
```