As data source, we use the Dataset from the "Happywhale - Whale and Dolphin Identification" competition from Kaggle. The goal is to identify new images from whales and assigning them to existing or identifying them as new whales.
In order to install all necessary dependencies, please make sure you have python3 and a local Conda distribution (e.g. anaconda or miniconda) installed.
After you cloned this repo, you can install the necessary dependencies via pip install -r requirements.txt.
You can download the original data from Kaggle but you have to make sure that you have enough free space on your disk.
After that you can run crop_and_save_images.py to downsize the images. Please make sure that you name the directory that you downloaded from Kaggle like this:KaggleData.
If you want to save time, you can also download our already preprocessed images and csv files from Google Drive. Everything you need is stored in OurTrainingData.zip. Make sure to unzip the file in the same parent directory as this repository.
You can also find useful logs and nice visualisationsfrom our training process on Google Drive!
src - python directory containing all of our main code
visualizations - containing visualisation & data analysis notebooks of the training dataset and really nice visaulisation of our training process, have a look at them!
training - containing two examplitory training pipeline notebooks