Multiscale Deep Network with Centerness-Aware Loss for Salient Object Detection

Table 1

Detailed parameters of three different operators, including convolutional operator, batch normalization operator, and ReLU operator, respectively. The left two parameters of Conv2d are the number of input and output channels, and the right two parameters are the kernel size. The parameter of the BatchNorm2d operator is the number of feature channels, and the parameter of the ReLU operator is in place.

No.	Feature	Conv2d	BatchNorm2d	ReLU	Result
1	F^1	2048, 512, 1 × 1	512	True	T^1
2	F^2	1024, 256, 1 × 1	256	True	T^2
3	F^3	512, 128, 1 × 1	128	True	T^3
4	F^4	256, 64, 1 × 1	64	True	T^4
5	F^5	128, 32, 1 × 1	32	True	T^5