Skip to content

Latest commit

 

History

History
32 lines (24 loc) · 2.32 KB

README.md

File metadata and controls

32 lines (24 loc) · 2.32 KB
Raw

RetinaNet OBB

Introduction

The RetinaNet of orietned form. Deriectly treat the horizontal anchors as oriented anchors with 0 theta. Use OBBOverlaps to calculator IoU when assigning targets.

@inproceedings{lin2017focal,
  title={Focal loss for dense object detection},
  author={Lin, Tsung-Yi and Goyal, Priya and Girshick, Ross and He, Kaiming and Doll{\'a}r, Piotr},
  booktitle={Proceedings of the IEEE international conference on computer vision},
  year={2017}
}

Results and Models

DOTA1.0

Backbone Data Lr schd Ms Rotate box AP Baidu Yun Google Drive
R50-FPN DOTA1.0 1x - - 69.4 key: wqrb model
R50-FPN DOTA1.0 2x - - 72.7 key: x41s -
R101-FPN DOTA1.0 1x - - 70.6 key: f5k6 -
R101-FPN DOTA1.0 2x - - 73.7 key: v8r9 -

HRSC

Backbone Data Lr schd Ms Rotate box AP Baidu Yun Google Drive
R50-FPN HRSC 1x - - 73.1 key: y5j8 -
R50-FPN HRSC 1x - - 73.3 key: aew1 -