Table 1.

Comparison with the state-of-the-art results on MS-COCO, symbol “*” means our re-implemented results.

Method	Backbone	Epoch	AP	${\mathit{AP}}_{50}$	${\mathit{AP}}_{75}$	${\mathit{AP}}_{\mathit{S}}$	${\mathit{AP}}_{\mathit{M}}$	${\mathit{AP}}_{\mathit{L}}$
One-stage
FoveaBox * [14]	ResNet50	1×	36.2	56.1	38.5	20.4	39.7	46.2
FoveaBox * [14]	ResNet101	1×	38.3	58.3	40.9	21.4	42.4	50.0
FoveaBox * [14]	ResNet101	2×	38.9	58.7	42.0	22.1	42.8	50.6
YOLOV2 [33]	Darknet19	-	21.6	44.0	19.2	5.0	22.4	35.5
YOLOV3 [10]	Darknet53	-	33.0	57.9	34.4	18.3	35.4	41.9
RetinaNet [34]	ResNet101	-	39.1	59.1	42.3	21.8	42.7	50.2
FCOS * [35]	ResNet50	1×	37.0	56.6	39.4	20.8	39.8	46.4
SSD512 [13]	ResNet101	-	31.2	50.4	33.3	10.2	34.5	49.8
CARAFE [36]	ResNet50	-	38.1	60.7	41.0	22.8	41.2	46.9
Two-stage
Faster RCNN * [7]	ResNet50	1×	36.5	58.7	39.1	21.5	39.7	44.6
Faster RCNN * [7]	ResNet101	1×	38.9	60.9	42.3	22.4	42.4	48.3
Faster RCNN * [7]	ResNet101	2×	39.7	61.4	43.3	22.3	42.9	50.4
ours
FoveaBox w/SEFPN	ResNet50	1×	37.3	58.0	39.6	22.0	41.0	47.7
FoveaBox w/SEFPN	ResNet101	1×	39.1	59.5	41.7	22.9	43.0	50.6
FoveaBox w/SEFPN	ResNet101	2×	39.9	61.0	42.8	23.3	43.8	52.6
Faster RCNN w/SEFPN	ResNet50	1×	37.3	58.6	40.6	22.2	40.3	47.7
Faster RCNN w/SEFPN	ResNet101	1×	39.6	60.6	43.4	22.8	43.5	52.0
Faster RCNN w/SEFPN	ResNet101	2×	39.7	60.2	43.6	22.3	43.2	52.5