Why the padding of first conv layer in FCN32s is 100?

[Mao Shuyi]

I wonder why the padding of first conv layer in FCN32s is 100?
FCN32s(
(conv1_1): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(100, 100))
I am a newcomer in ML.Can you give me some answer? Thank you for advance

[Yuwen Heng]

It's from the original code at https://github.com/shelhamer/fcn.berkeleyvision.org/blob/master/voc-fcn16s/net.py

[Mao Shuyi]

It's from the original code at https://github.com/shelhamer/fcn.berkeleyvision.org/blob/master/voc-fcn16s/net.py

emm.....I still don't understand why the padding of first conv layer in FCN32s is 100?

[Yuwen Heng]

To make sure the network can take input with any size. The pooling layer will down sample the input so there is a limit of the input sample size. With padding, we can take inputs of any size, for example, 16*16 image. At the output, we crop the desired part which has the same shape as input.