Run pip install -r requirements.txt to install all the dependencies.
From the cloned repo, first run ./data/get_data.sh and then ./setup to make the data directory compatible with custom dataloaders.
Run python test_submission.py [eval.csv] where [eval.csv] is your file with columns: Image_id (int), image_path (str), image_height (int), image_width (int), image_channels (int). This will produce a file eval_classified.csv containing an id and predicted class for each image.