To help speed up research on cell microscopy images, this semantic segmentation program creates masks of the cell from an image of a cell.
Create conda environment using the environment.yml file and using the command: conda env create -f environment.yml
If you don't have conda or conda isn't working run:
pip install -r requirements.txt
edit path in Train_model.py make sure to change training and validate directory in Train_model.py
Then just run python Train_model.py
First, install wandb pip install wandb (if not installed)
I use specified_gpu_wandb.py the most. I just set the GPU at the bottom
of the file.
wandb_train.py runs with any available gpu.
Make new sweep: wandb sweep wandb_unet.yaml Start agent: wandb agent // (should be able to copy and past from the dashboard info or after you make the sweep)
In the wandb_unet.yaml file is an example of how to set what parameters to test.
There are 3 testing files currently in use:
edit "root = "TrainingDataset/data_subset/output/test/"" with a path to
a directory of images. (line 41)
- *Test_patched_predictions - show full size image, label, and prediction (I use this one the most)
- Test_model.py - show image, label patch, predication patch
- Test_full_image - show each prediction patch, as well as full size image, label, and prediction
Make sure to change root and test variables to the correct directories
Segment_timelapse.py segments each image in a directory and saves to another specified directory.
Usage:
Segment_timelapse.py -i "input_folder" -o "output_folder" --confidence 0.5 --see_confidence_mask False
- i: path to input folder
- o: path to output folder
- c: confidence level between 0.0 and 1.0 (higher under segments, lower over segments)
- s: see confidence mask - if this is set to False you get a binary mask, True overrides confidence level and produces a mask of confidence levels for each pixel.
*Windows doesn't work with ray. I can setup Windows support if there is a need.