ASBL-RetinaNet
This repository provide the code of Adaptive Saliency Biased Loss (ASBL). We provide model weights for each dataset shown in the paper for replication of our work. In addition, basic code structure provided in this repo. This repo is currently in maintainance. If you have any question when you run the code, please feel free to leave message in the issues channel, we will keep it posted.
DOTA weight: https://drive.google.com/open?id=1Fpk_E5l7340Z1JF1akeSDRRmFMUOxKNK
NWPU VHR-10 weight: https://drive.google.com/open?id=1dokCVdC8NXGNHmt0vURa4rOVYE6IBcCA
VGG-16 Pretrained weight: https://drive.google.com/open?id=1IVnDC6oWmWnBvl7COQ_qgZAyvcdnlSwB
ResNet50 Pretrained weight: https://drive.google.com/open?id=1_2QMI7mbZuUvhIenW_3-MctF6Ut3WmGg
test inference
Enviroment
pytorch = 0.3.1