Use Depthwise Convolution instead of Group Convolution to accelerate training MobileNet in Caffe.
Primary LanguageCuda