---
license: apache-2.0
task_categories:
- image-segmentation
tags:
- medical
- biology
---

# ACDC-PNG Dataset

[Paper](https://arxiv.org/abs/2601.10124) | [Code](https://github.com/script-Yang/VQ-Seg)

This repository contains a convenient PNG-formatted version of the ACDC dataset, primarily intended for semi-supervised medical image segmentation tasks. This version was converted from the files provided in the [SSL4MIS repository](https://github.com/HiLab-git/SSL4MIS/tree/master/data/ACDC).

It is used and introduced in the paper: **VQ-Seg: Vector-Quantized Token Perturbation for Semi-Supervised Medical Image Segmentation**.

### Dataset Structure

The data is organized as follows:

```bash
XXX/
│
├── train-label/                # Labeled training set
│   ├── image/                  # Input images (.png)
│   └── mask/                   # Corresponding segmentation masks (.png)
│
├── train-unlabel/              # Unlabeled training set
│   └── image/                  # Images without ground truth masks
│   └── mask/                  
│
├── val/                        # Validation set
│   ├── image/                  # Validation images (.png)
│   └── mask/                   # Validation masks (.png)
│
└── test/                       # Test set
    ├── image/                  # Test images (.png)
    └── mask/                   # Test masks (optional)
```

- **train-label**: Paired image–mask samples used for supervised segmentation training.
- **train-unlabel**: Images without ground-truth annotations, utilized for semi-supervised learning.
- **val**: Used to monitor and validate model performance during training.
- **test**: Used for final evaluation and benchmarking.

### Citation

If you use this dataset or the VQ-Seg method in your research, please cite the following:

```bibtex
@inproceedings{yangvq,
  title={VQ-Seg: Vector-Quantized Token Perturbation for Semi-Supervised Medical Image Segmentation},
  author={Yang, Sicheng and Xing, Zhaohu and Zhu, Lei},
  booktitle={The Thirty-ninth Annual Conference on Neural Information Processing Systems}
}
```

Please also make sure to cite the **original ACDC paper**.