Details on Inv Depth Layer and ConstantPad2D

[F-Barto]

Hello,

I would like to know why are you explicitly using ConstantPad2D ?

[link to code]

self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=3, stride=1)
self.pad = nn.ConstantPad2d([1] * 4, value=0)

From the doc nn.Conv2d default padding_mode is to zeros. Isn't setting padding=1 in Conv2d equivalent to your self.conv1(self.pad(x)) ?

Am I missing something ?

Have a good day