First download the CAR-A and CAR-B dataset from http://www.orand.cl/en/icfhr2014-hdsr/#datasets.
Or use your own dataset and adapt its structure to match that of the CAR datasets.
The training can be started by running the train.py
script. For example
python train.py --data '/path/to/ORAND-CAR-2014/CAR-A/' --epochs 20
Check python train.py -h
for all possible options.
TODO