Hi,
I have a bit tricky question. I am training a PixelCNN network that I am working on. In my data, there are some points that I don’t want my CNN to take in account. First I applied a Mask on my data and it works for some points but for others it is not working. Then I found the problem, at those points, when the kernel (3 X 3) convoluted, it is taking the account of the data points which are unmasked or necessary surrounding points. For example in the data below, for the masked point “A” the kernel will take account for two 1s in the last row of first matrix.
[[[[ 0., 0., 0., 0.],
[ A., 0., 0., 0.],
[ 0., 0., 0., 0.],
[ 0., 1., 1., 1.]],
[[ 0., 0., 0., 0.],
[-1., 1., -1., -1.],
[-1., 1., 1., 1.],
[ 1., -1., 1., 0.]]]], device='cuda:1')
My question is, is there anything that I can do, so that my kernel will skip these points (for example “A”, or the 0s in above dataset) or I can do something else? Masking them won’t help and I can’t remove these 0s from my dataset.
Thank you