atharvacc / MachineChurning

AI BASED COLOR MAPPER FOR Microscopy

We provide 2 models and a parser to train the model

• Cycle-Gan with saliency loss
• Gray-scale converter

Installing the libraries

pip install -r requirements.txt
#Run the command below if you get GPU Driver issues
pip install torch==1.5.0+cu101 torchvision==0.6.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html

Training the model and inference

• pretrained models can be found here https://drive.google.com/drive/folders/1FdxMdDVCZHN-V0x_tOvS4wemccTl6XEg?usp=sharing
• Copy the pretrained model folder to MachineChurning/App/. THE DIRECTORY STRUCTURE IS IMPORTANT FOR THE CODE TO BUILD.
• Copy your files to the MachineChurning Directory with the following tree structure

MUSE/
├── testA
├── testB
├── trainA
└── trainB

python main_parser.py --train_or_test train <args> # For training
python main_parser.py --train_or_test test <args> # For testing
# Examples for training cycle-gan can be found in scripts/train_script.sh

The documentation for main_parser can be found under docs/parser_documentation.md

Loading the interface

• Navigate to App
• Run the commands below

bash run_flask.sh #Run on port 8888
bash run_flask.sh local #Run locally

A few results

Original MUSE

Predicted H&E