ISBI2012 Segmentation with PyTorch

Basic Info

Dataset

ISBI2012: http://brainiac2.mit.edu/isbi_challenge

Backbone Network

U-Net: Convolutional Networks for Biomedical Image Segmentation \<arxiv link>

@misc{ronneberger2015unet,
    title={U-Net: Convolutional Networks for Biomedical Image Segmentation},
    author={Olaf Ronneberger and Philipp Fischer and Thomas Brox},
    year={2015},
    eprint={1505.04597},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}

Data Preprocessing

• Download ISBI2012 Dataset from http://brainiac2.mit.edu/isbi_challenge
• Save data like below.

# ROOT Directory
ISBI2012
├── train-volume.tif
├── train-labels.tif
└── test-volume.tif --> (Not Necessary)

• Split Data to train, val, test with data_prep.ipynb

Train

$ python train.py

Test

$ python eval.py

Then, you can see result images from ./test_results

Reference

• hanyoseob - youtube-cnn-002-pytorch-unet