ziiio/SketchDUO

---
pretty_name: SketchDUO
homepage: https://zihos.github.io/StableSketcher
language:
- en
license: other
size_categories:
- 1K<n<100K
task_categories:
- image-text-to-text
- visual-question-answering
tags:
- image
- text
- multimodal
- captions
- vqa
---

# SketchDUO

This dataset contains sketch images with optional captions and optional question-answer pairs.

## Structure

- Splits: `positive`, `negative`
- Each row contains:
- `image`
- `caption`
- `qa_pairs`
- `has_caption`
- `has_qa`

## Notes

- All image files under each split are uploaded as rows.
- When a caption or QA annotation is missing for an image, the corresponding field is left empty.
- Rows without captions correspond to augmented data.
- QA pairs are grouped per image into the `qa_pairs` column.

## Counts

- Total rows: 35851
- Rows with captions: 4693
- Rows with QA: 4692
- Image-only rows: 31158

## Split counts

- `positive`: 24000 rows
- `negative`: 11851 rows

## Repository

- Hub repo: `ziiio/SketchDUO`
- Project page: `https://zihos.github.io/StableSketcher`
- Paper: `https://arxiv.org/abs/2510.20093`

## Citation

```bibtex
@article{park2025stablesketcher,
title={StableSketcher: Enhancing Diffusion Model for Pixel-based Sketch Generation via Visual Question Answering Feedback},
author={Park, Jiho and Choi, Sieun and Seo, Jaeyoon and Kim, Jihie},
journal={arXiv preprint arXiv:2510.20093},
year={2025}
}
```