group conv layer

[runningJ]

can you share group conv layer?

[zunzhumu]

@runningJ Sorry. You can try to complete the group conv layer by caffe code. Gook luck!

[fishman2008]

I follow the caffe implementation and put a for.loop outside of gemm. I have properly checked the value of weight_offset, col_offset, and output_offset. Nothing is modified. The group value is set to be l.c, known as input channel number. I got the access violation error. Could you provide any guidelines? Thanks. what else should I change?

[zunzhumu]

@fishman2008 I changed the size of l.weights of convolution layer (Turn l.cl.nl.sizel.size into l.cl.nl.sizel.size/l.group in convolutional_layer.c of make_convolutional_layer function ). I'm not sure where your mistake is, unless it is provided provide the picture of your error. Good Luck!