Pytorch implementation of RetinaNet object detection as described in Focal Loss for Dense Object Detection by Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He and Piotr Dollár.
Along with additions form Sorting out Lipschitz function approximation to ensure the network is lipschitz continuous.
This implementation is primarily designed to be easy to read and simple to modify.
Working on it
- Significant amounts of code are borrowed from the keras retinanet implementation
- The NMS module used is from the pytorch faster-rcnn implementation
- The implementations of Group Sort, and the Bjorck Convolution is from the LNets