hongyanggao / channelnets Goto Github PK

Hello, I am very interested in your work, and I would like to repeat your results.How do I use your code？

您好，channel-wise convolution的实现方式是utils/ops.py文件里面的rev_conv2d函数么？
如果是这样，似乎实现方式有点不对。conv1d应该在[-1, num_input_channels x h x w,1]形状的数据上做卷积，而不是在[-1, num_input_channels，h x w]上做卷积。

I‘ve read your paper, it's a nice work. I notice you mention MobileNet_v2 in your paper, but in the comparison between ChannelNet-v1 and other CNNs in terms of the top-1 accuracy on the ImageNet validation set, it isn't include the result of mobilenet_v2. I wonder if you have the plan to improve channelnet with inverted residual block based on mobilenet_v2?

Could you provide the FLOPs of ChannelNets-v2 and v3?

Can you provide the trained-model? thank you!