pytorch_compact_bilinear_pooling
pytorch_compact_bilinear_pooling copied to clipboard
gdlg
Dimension problem
Dear gdlg, Wishing you all the best lately. Thank you very much for sharing the CBP codes. The time I was trying to test them, I found an interesting result. My test codes are shown as follows: import torch from compact_bilinear_pooling import CountSketch, CompactBilinearPooling
input_size = 3 output_size = 3 mcb = CompactBilinearPooling(input_size, input_size, output_size).cuda() x = torch.rand(4,32,32,3).cuda() y = torch.rand(4,32,32,3).cuda()
z = mcb(x,y) As you had answered in another issues named "Merge"，we should make sure the channel dimension be put in the last dimension of tensor, I had follow this advise in the code. With the codes above, I expected to get the output z with dimension (4, 32, 32, 3), however, the actual output dimension of z is (4, 3, 32, 2). I have been confused about this problem for a while. It would be very kind of you to take some time out of your busy schedule to discuss this issue with me.
