Table 1.

Double-branch network structure.

Layer name	Layer type	Classification branch	Regression branch
L1	Conv	(3 × 3,32) × 3(=7 × 7)
		2 × 1 max pooling
L2	Conv	(3 × 3,64) × 2(=5 × 5)
		2 × 1 max pooling
L3	Conv	(3 × 3,128) × 2(=5 × 5)
		2 × 2 max pooling	2 × 1 max pooling
L4	Conv	(3 × 3,256)	(3 × 3,256)
		2 × 2 max pooling	2 × 1 max pooling
L5	Conv	(3 × 3,512)	(3 × 3,512)
		2 × 2 max pooling	2 × 1 max pooling
L6	Conv	(3 × 3,512)	(1 × 1,256)
		GAP	(1 × 1,1)
L7	Output	Output 1 (size 2 × 1)	Output 2 (size 720 × 1)