Fusion Attention Mechanism for Foreground Detection Based on Multiscale U-Net Architecture

Table 1

The proposed AMU-Net model configurations.

Layer	Kernel	Stride	Channel	Output size
Input	—	—	3
conv1_1		1	64
conv1_2		1	64
maxpool_1		2	64
conv2_1		1	128
conv2_2		1	128
maxpool_2		2	128
conv3_1		1	256
conv3_2		1	256
conv3_3		1	256
maxpool_3		2	256
conv4_1		1	512
conv4_2		1	512
conv4_3		1	512
maxpool_4		2	512
conv5_1		1	512
conv5_2		1	512
conv5_3		1	512
conv6		1	512
attention4		1	512
attention3		1	256
attention2		1	128
attention1		1	64
tranconv4		2	256
conv4d		1	512
tranconv3		2	256
conv3d		1	256
tranconv2		2	128
conv2d		1	128
tranconv1		2	64
conv1d		1	64
conv_out		1	1