An Artificial Intelligence assisted tool for automated microscopy.
git clone https://github.com/ScotfaAI/CelFDrive.git
conda env create -f environment-gpu-windows.yml --prefix Path/to/env
Edit the global to set the absolute location of your cloned repository.
repo_path = "path\\to\\CelFDrive"
The primary function in predict.py is get_target_location it takes:
To modify this to work for a problem outside mitosis the get_target_location function, the class_info dictionary needs to be modified to your experiment.
class_info = {
class_id: (class_name, acceptable_confidence, priority_ranking),
...
}
If you have another model you can replace Ultralytics with a pytorch model, but you may need to modify the coordinate calculations as the current version is based on normalised xywh coordinates where xy is the centre of the object.
CellClicker and CellSelector can be used to generate YOLO compatible labels from time series data.
Each dataset for training needs to contained in a single folder which has a subfolder called "images". Each file should end in t001.extension for a given timepoint.
From inside your conda environment:
python run_clicker.py
Edit run_selector.py and change this line to be all of your ordered classes.
phases = ['prophase','earlyprometaphase', 'prometaphase', 'metaphase', 'anaphase', 'telophase']
From inside your conda environment:
python run_selector.py
Edit the following variables in run_conversion.py
user = 'Scott'
imgpath = 'Path/to/Dataset'
From inside your conda environment:
python run_conversion.py
This software is easy to deploy with intelligent-imaging-innovations Conditional Capture, but will work with imaging software that allows python code to be run such as LabVIEW and Micro-Manager.