This repository provides the source code of the Deep Learning model used in the manuscript "Genetic dissection of dehulling efficiency in sunflower". Here, we trained and evaluated a Convolutional Neural Networks to predict the dehulling efficiency based on RGB images.
| Metric | Value |
|---|---|
| F1-Score | 0.37 |
| Acc. with tolerance = 1 | 0.84 |
| MAE | 0.80 |
| Cohen's Quadratic Kappa | 0.84 |
- Clone this project
git clone git@github.com:grimmlab/SunflowerDehullingEfficiency.git
- Install requirements
cd SunflowerDehullingEfficiency
pip install -r requirements.txt
- Download the folder with the dataset split files (
splits) from Mendeley Data (doi.org/10.17632/27d9gfczt3.1) and paste it inSunflowerDehullingEfficiency/data. - Download image patches (
patches.zip) from Mendeley Data and paste the unzipped folder inSunflowerDehullingEfficiency/data. - Download the folder with the model file (
models.zip) from Mendeley Data and paste the unzipped folder inSunflowerDehullingEfficiency. - Run inference on the test set. This step will print the results to the console.
python3 run_inference.py
Hyperparameters can be optimized using the following command. Change the file according to your use-case.
python3 optimize_hyperparameter_focal_loss.py