https://github.com/hust-linyi/insmix

https://github.com/hust-linyi/insmix

Last synced: about 2 months ago
JSON representation

• Host: GitHub
• URL: https://github.com/hust-linyi/insmix
• Owner: hust-linyi
• Created: 2022-06-15T08:15:18.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2023-03-16T17:13:26.000Z (almost 2 years ago)
• Last Synced: 2024-08-24T17:26:17.656Z (6 months ago)
• Language: Python
• Size: 23.5 MB
• Stars: 8
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• Awesome-Mixup - [Code

README

        
# InsMix

This is the official code for "InsMix: Towards Realistic Generative Data Augmentation for Nuclei Instance Segmentation (MICCAI 2022, early accepted)"

### Pipeline

![pipeline](figs/insmix1.png)

### Method

![method](figs/insmix2.png)

### Requirements

torch>=1.4.0

torchvision>=0.5.0

dominate>=2.4.0

visdom>=0.1.8.8

wandb

### Usage

#### InsMix w/o Smooth-GAN

The fuctions 'insmix' and 'background shuffle' can be found in 'data_aug.py'.

The example code for dataloader is in 'dataset.py'. Note that it can be used to [BRPNet](https://github.com/csccsccsccsc/brpnet) and [NB-Net](https://github.com/easycui/nuclei_segmentation), which utilize two types of label, i.e., the inner area and the boundary.

#### InsMix w/ Smooth-GAN

You may simply run the scripts as:

```

bash Smooth-GAN/scripts/train_nuclei.sh

bash Smooth-GAN/scripts/test_nuclei.sh

```

### Citation

Pleae cite the paper if you use the code.

```

@inproceedings{lin2022insmix,

  title={{InsMix}: Towards Realistic Generative Data Augmentation for Nuclei Instance Segmentation},

  author={Lin, Yi and Wang, Zeyu and Cheng, Kwang-Ting and Chen, Hao},

  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},

  year={2022},

  organization={Springer}

}

```

### TODO

- [ ] Training and testing on Kumar dataset.

- [ ] Refactor the code to make it more readable.

### Acknowledgment 

The code of Smooth-GAN is heavily build on [pix2pix](https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix), thanks for their amazing work!