---
tags:
- image-stitching
pretty_name: UniStitch_Datasets
size_categories:
- 100M<n<1B
---
# **UniStitch: Unifying Semantic and Geometric Features for Image Stitching**

<p align="center">
      &nbsp&nbsp 🖥️ <a href="https://github.com/MmelodYy/UniStitch">GitHub</a> &nbsp&nbsp  | &nbsp&nbsp🤗 <a href="https://huggingface.co/Y5Y/UniStitch_model">Hugging Face Model</a>&nbsp&nbsp | &nbsp&nbsp 📑 <a href="https://arxiv.org/abs/2603.10568">Paper </a> &nbsp&nbsp
<br>


## Dataset Details

The dataset provided here mainly consists of keypoint features extracted for the **UDIS** and **classical** datasets used in the Unistitch paper. You can directly place the **superpoint_lmdb** folder corresponding to this dataset into the folder of the respective sub-dataset, ensuring it is at the same level as **input1** and **input2**.

| | |
|---|---|
| **Developed by** | Yuan Mei, Lang Nie, Kang Liao, Yunqiu Xu, Chunyu Lin, Bin Xiao |
| **Affiliation** | CQUPT, PolyU, NTU, NUS, BJTU |
| **First released** | arXiv pre-print, 2026 |


### Citation
If you find Unistitch useful for your research or applications, please cite our paper using the following BibTeX:

```bibtex
  @article{mei2026unistitch,
  title={UniStitch: Unifying Semantic and Geometric Features for Image Stitching},
  author={Mei, Yuan and Nie, Lang and Liao, Kang and Xu, Yunqiu and Lin, Chunyu and Xiao, Bin},
  journal={arXiv preprint arXiv:2603.10568},
  year={2026}
}
```