What is the difference between binary_conv_net.py and binary_net.py?
Is the xnor operation done on the convolution layers in such a way that, matrix elements work with xnor operation and pop-count operation? We can find that, conv2d operator is not changed, from the base. So, will it do normal matrix multiplication with binarized elements, or will it do the xnor operation on the same? Specifically alpha operator will be used in the xnor-net implementation during the 1x1 pointwise convolution.. Where exactly does this operator occur in the implementation?
What is the difference between binary_conv_net.py and binary_net.py?
Is the xnor operation done on the convolution layers in such a way that, matrix elements work with xnor operation and pop-count operation? We can find that, conv2d operator is not changed, from the base. So, will it do normal matrix multiplication with binarized elements, or will it do the xnor operation on the same? Specifically alpha operator will be used in the xnor-net implementation during the 1x1 pointwise convolution.. Where exactly does this operator occur in the implementation?