Keras implementation of RetinaNet object detection forked from the original one (see link above) described in Focal Loss for Dense Object Detection by Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He and Piotr Dollár.

This version was modified in order to fit the requirements of the problem:

• source data is stored in numpy arrays
• source data along with the labels were augmented in a particular way (see preprocessing repository here)
• satellite mosaics are too large, so cropping was involved at the preprocessing stage