Skip to main content
. 2020 Feb 3;20(3):812. doi: 10.3390/s20030812

Figure 3.

Figure 3

De-convolution with the full mode (N = 4, M = 3 in this example). The size of the input feature map is N × N. We expand the original feature maps with M − 1 paddings (padding pixel = 0) around. Then, we use an M × M kernel to carry out the convolutional operation with the “full” mode (padding = M − 1 and step = 1), which could obtain an (M + N − 1) × (M + N − 1) output.