Activate virtual environment:
python -m venv venv
source venv/bin/activate
Install libraries:
pip install -r requirements.txt
- https://www.robots.ox.ac.uk/~vgg/data/fgvc-aircraft/archives/fgvc-aircraft-2013b.tar.gz
- https://www.kaggle.com/sripaadsrinivasan/fish-species-image-data
- https://www.kaggle.com/ryanpartridge01/wild-edible-plants
- https://www.kaggle.com/gpiosenka/100-bird-species
- Cifar is downloaded through the script
To run training on all datasets and color models run the main.py script, or to run only one dataset use the corresponding script named after the daset, e.g. fish.py.
main.py does 5 runs of 25 epochs for each combination, while the individual scripts are conmfigured to do one run of 10 epochs for each color model.
We have also included a few utility scripts used for analysing the datasets and our results. These are dataset_averages.py, averages.py and plotter.py.
averages.py combines the csv files from multiple runs into a average csv, and plotter.py creates png plots from csv files.
dataset_averages.py is used to compute some statistics of the datasets, namely the mean, average and variance of each color channel (for all color models).