Improved stereo matching algorithm based on multi-scale fusion | Xibei Gongye Daxue Xuebao/Journal of Northwestern Polytechnical University

Open Access

表1

FS-CNN网络结构参数表

网络模块	输入	参数	输出	输入尺寸	输出尺寸
主干网络特征提取	left/right		input	15×15×1	15×15×1
	Input	K=4×4, C=128, S=1, BN, ReLU	Conv_0	15×15×1	12×12×128
	Conv_0	K=3×3, C=64, S=2, BN, ReLU	Conv_1	12×12×128	6×6×64
	Conv_1	K=3×3, C=32, S=2, BN, ReLU	Conv_2	6×6×64	3×3×32
	Conv_2	K=3×3, C=32, S=2, BN, ReLU	Conv_3	3×3×32	1×1×32
FPN结构特征融合	Conv_3	S=3×3	UPSampling_0(上采样)	1×1×32	3×3×32
	UPSampling_0, Conv_2		Add_0(特征融合)	3×3×32	3×3×32
	Add_0	S=2×2	UPSampling_1(上采样)	3×3×32	6×6×32
	Conv_1	K=1×1, C=32, S=1, ReLU	Conv_1_1(通道降维)	6×6×64	6×6×32
	UPSampling_1, Conv_1_1		Add_1(特征融合)	6×6×32	6×6×32
	Add_1	S=2×2	UPSampling_2(上采样)	6×6×32	12×12×32
	Conv_0	K=1×1, C=32, S=1, ReLU	Conv_0_0(通道降维)	12×12×128	12×12×32
	UPSampling_2, Conv_0_0		Add_2(特征融合)	12×12×32	12×12×32
特征匹配	Add_0~2	K=3×3, C=32, S=1, ReLU	Conv_Add _0~2	3/6/12×3/6/12×32	3/6/12×3/6/12×32
	Conv_Add _0~2		Flatten_0~2	3/6/12×3/6/12×32	288/1152/4608
	Flatten_0~2	S=300, ReLU	Dense0~2	288/1152/4608	300
	Dense0~2left, Dense0~2right		Concat_0~2	300	600
	Concat_0~2	S=521, ReLU	Dense0_0~2	600	521
	Dense0_0~2	S=256, ReLU	Dense1_0~2	521	256
	Dense1_0~2	S=2, Softmax	s_(1~3)	256	2

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.