FCOS: Fully Convolutional One-Stage Object Detection

Introduction

@article{tian2019fcos,
  title={FCOS: Fully Convolutional One-Stage Object Detection},
  author={Tian, Zhi and Shen, Chunhua and Chen, Hao and He, Tong},
  journal={arXiv preprint arXiv:1904.01355},
  year={2019}
}

Results and Models

Backbone	Style	GN	MS train	Tricks	DCN	Lr schd	Mem (GB)	Inf time (fps)	box AP	Config	Download
R-50	caffe	N	N	N	N	1x	5.2	22.9	36.2	config	model | log
R-50	caffe	Y	N	N	N	1x	6.5	22.7	36.6	config	model | log
R-50	caffe	Y	N	Y	N	1x	-	-	38.6	config	model | log
R-50	caffe	Y	N	Y	Y	1x	-	-	42.5	config	model | log
R-50	caffe	Y	N	N	N	2x	-	-	36.9	config	model | log
R-101	caffe	Y	N	N	N	1x	10.2	17.3	39.2	config	model | log
R-101	caffe	Y	N	N	N	2x	-	-	39.1	config	model | log

Backbone	Style	GN	MS train	Lr schd	Mem (GB)	Inf time (fps)	box AP	Config	Download
R-50	caffe	Y	Y	2x	6.5	22.9	38.7
R-101	caffe	Y	Y	2x	10.2	17.3	40.9	config	model | log
X-101	pytorch	Y	Y	2x	10.0	9.3	42.5	config	model | log

Notes:

• To be consistent with the author's implementation, we use 4 GPUs with 4 images/GPU for R-50 and R-101 models, and 8 GPUs with 2 image/GPU for X-101 models.
• The X-101 backbone is X-101-64x4d.
• Tricks means setting norm_on_bbox, centerness_on_reg, center_sampling as True.
• DCN means using DCNv2 in both backbone and head.