My Final Year Project(FYP) In National University of Singapore(NUS)

You need

Pytorch(stable 1.9.1) 

Both cuda version and cpu version are OK

File Structure

📦FYP-U-Net
 ┣ 📂data
 ┃ ┣ 📂imgs
 ┃ ┃ ┣ 📌···.tif
 ┃ ┃ ┗ ···
 ┃ ┣ 📂masks
 ┃ ┃ ┣ 📌···_mask.tif
 ┃ ┃ ┗ ···
 ┃ ┣ 📂PredictImage 
 ┃ ┃ ┣ 📌0.tif
 ┃ ┃ ┣ 📌1.tif
 ┃ ┃ ┗ ···
 ┃ ┣ 📂SaveImage
 ┃ ┃ ┣ 📌0.tif
 ┃ ┃ ┣ 📌1.tif
 ┃ ┃ ┗ ···
 ┃ ┗ 📂Source
 ┃ ┃ ┣ 📂TCGA_CS_4941_19960909
 ┃ ┃ ┃ ┣ 📌TCGA_CS_4941_19960909_1.tif
 ┃ ┃ ┃ ┣ 📌TCGA_CS_4941_19960909_1_mask.tif 
 ┃ ┃ ┃ ┣ 📌TCGA_CS_4941_19960909_2.tif
 ┃ ┃ ┃ ┣ 📌TCGA_CS_4941_19960909_2_mask.tif 
 ┃ ┃ ┃ ┗ ···
 ┃ ┃ ┣ 📂TCGA_CS_4942_19970222
 ┃ ┃ ┗ ···
 ┣ 📂params
 ┃ ┗ 📜unet.pth
 ┣ 📓README,md
 ┣ 📄data.py
 ┣ 📄net.py
 ┣ 📄utils.py
 ┗ 📄train.py

• 'data' dir contains the origin dataset in 'Source' dir. And the dataset can be download in Kaggle (https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/). And also you can use different dataset.
• 'imgs' contains images and 'masks' contains corresponding masks to images. Corresponding masks have a _mask suffix. More inforamtion you can check in kaggle.
• 'SaveImage' is meant for store train results and 'PredictImage' is meant for store test results.
• 'params' is meant for store model.

Quick Up

Run train.py

Change DataSet

• Delte all images in data dir and its subdir.

• Install dataset from kaggle or anything you like(PS. Corresponding masks must have a _mask suffix) into 'Source' dir

• Run data.py

  python3 data.py
  

  Remember change the path. After this, you will get images and masks in imgs dir and masks dir.

• Run train.py

  python3 train.py
  

  Remember change the path. And you can see the results in 'SaveImage' dir and 'PredictImage' dir.

Results

Pre-trained model

https://drive.google.com/file/d/1yyrITv7BQf9kDnP__g6Qa3_wUPD1c_i_/view?usp=sharing