A Small Object Detection Network Based on Multiple Feature Enhancement and Feature Fusion

Table 2

Effect of each component on COCO val-2017.

RetinaNet	BidiFPN	MCEM	SCEM	AP	AP50	AP75	APS	APM	APL
√	×	×	×	32.5	50.9	34.8	13.9	35.8	46.7
√	√	×	×	33.3	51.6	35.8	14.9	36.3	47.1
√	√	√	×	34.1	52.3	36.8	16.0	36.9	47.4
√	√	×	√	34.0	52.1	36.6	15.7	36.5	47.5
√	√	√	√	34.8	52.8	37.4	16.8	37.3	47.9

Note: BidiFPN: bidirectional fusion feature pyramid structure; MCEM: multiscale context extraction module; SCEM: subpixel convolution enhancement module.