3D Semantic VSLAM of Indoor Environment Based on Mask Scoring RCNN

Table 1

Comparative results of MS RCNN algorithm and other instance segment algorithms on COCO testing set.

Method	Backbone	AP	[email protected]	[email protected]	APS	APM	APL
MNC [27]	ResNet-101	23.2	43.2	25.1	4.5	24.8	44.3
FCIS [28]	ResNet-101	28.9	48.7	—	—	—	—
FCIS+++ [28]	ResNet-101	34.2	53.7	—	—	—	—
Mask RCNN [14]	ResNeXt-101 FPN	36.9	61.2	38.6	17.1	38.7	52.4
MaskLab+ [29]	ResNet-101(JET)	37.8	62.4	41.0	18.2	40.9	50.7
Mask RCNN	ResNet-101	33.3	55.0	36.6	13.2	36.4	52.3
MS RCNN		35.4	54.9	38.1	13.7	37.6	53.3
Mask RCNN	ResNet-101 FPN	37.0	59.2	39.5	17.1	39.3	52.9
MS RCNN		38.3	58.8	41.5	17.8	40.4	54.4
Mask RCNN	ResNet-101-DCN+FPN	38.4	61.2	41.2	18.0	40.5	55.2
MS RCNN		39.6	60.7	43.1	18.8	41.5	56.2