---
license: cc-by-4.0
task_categories:
  - image-to-image
tags:
  - polarization
  - linear-rgb
  - ipid
size_categories:
  - n<1K
---

# IPID

This release contains the images supplied in the WU folders of the local Linear_RGB_Polar RLP collection. Dataset and scene directory names have been changed from WU to IPID, while the original training and test split and PNG contents are preserved.

## Structure

```text
train/IPID001/{0,45,90,135}.png
...
test/IPID004/{0,45,90,135}.png
...
```

Each scene contains four 8-bit RGB PNG images named for polarization angles in degrees: 0, 45, 90, and 135. Images are uploaded without conversion.

| Split | Scenes | PNG files | Bytes |
| --- | ---: | ---: | ---: |
| train | 170 | 680 | 881328742 |
| test | 7 | 28 | 33856651 |

Scene identifiers range from IPID001 through IPID177. Test scenes are IPID004, IPID017, IPID046, IPID070, IPID085, IPID086, and IPID096; all other scenes belong to the training split.

Image dimensions: 1024 x 1024 (628 images), 1280 x 1280 (20), 1536 x 1536 (32), 1664 x 1664 (4), 1792 x 1792 (16), 2048 x 1536 (4), and 2448 x 2048 (4).

License: CC-BY-4.0.

The accompanying `manifest.json` records each image's repository path, byte size, and SHA-256 checksum.

## Citation

If you use this dataset in your research, please cite:

```bibtex
@article{wu2026deep,
  title={Deep unfolding with learned regularization for color polarization demosaicking},
  author={Wu, Caiyun and Luo, Yidong and Li, Chenggong and Zhang, Junchao and Lianga, Buge},
  journal={Advanced Imaging},
  volume={3},
  number={3},
  year={2026},
  publisher={Science Press}
}
```