retinanet_obb

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	-