https://github.com/hubmapconsortium/pas-ftu-segmentation-pipeline

https://github.com/hubmapconsortium/pas-ftu-segmentation-pipeline

ot2od030545

Last synced: 17 days ago
JSON representation

• Host: GitHub
• URL: https://github.com/hubmapconsortium/pas-ftu-segmentation-pipeline
• Owner: hubmapconsortium
• License: mit
• Created: 2022-07-20T15:49:09.000Z (almost 4 years ago)
• Default Branch: main
• Last Pushed: 2026-03-02T14:57:40.000Z (3 months ago)
• Last Synced: 2026-03-02T17:59:40.911Z (3 months ago)
• Topics: ot2od030545
• Language: Python
• Homepage:
• Size: 382 KB
• Stars: 0
• Watchers: 6
• Forks: 0
• Open Issues: 3
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# 1-Tom

### Description

- The model uses a single U-Net SeResNext101 architecture with Convolutional Block Attention Module (CBAM), hypercolumns, and deep supervision. 

- It reads the WSIs as tiled 1024x1024 pixel images and then further resized as 320x320 tiles and sampled using a balanced sampling strategy. 

- The model is trained using a combination of Binary Cross-entropy loss and Lovász Hinge loss, and the optimizer used is SGD (Stochastic gradient descent). Training is for 20 epochs, with a learning rate of 1e-4 to 1e-6 and batch size of 8.

- For the model trained on colon data from scratch or using transfer learning, the training is done for 50-100 epochs and the validation set is increased from 1 slide to 2 slides.

### Requirements

- albumentations

- opencv-python

- pandas

- torch

- tqdm

- Python 3

- CUDA 

- cuddn 

- nvidia drivers 

- See `kaggle-hubmap-main/requirements.txt` file for a detailed list of dependencies.

### Usage

- Use Inference.py to run inference on a dataset.

- Use train folder to train on a dataset.

    cd src

    cd 01_data_preparation/01_01

    python data_preparation_01_01.py

    cd ..

    cd 01_02

    python data_preparation_01_01.py

    cd ..

    cd ..

    cd 02_train

    python train_02.py