Caffe Group parameter

The Convolution_param parameter Dictionary of the Caffe convolution layer has a group parameter, which means that the corresponding input channel is grouped with the output channel number, for example, the input data size is
90X100X100X32 90 is the size of the data batch 100x100 is the image data shape,32 is the number of channels, to go through a 3x3x48 convolution, group default is 1, is fully connected to the convolution layer,
If the group is 2, then the 32 channels to be entered are divided into 2 16 channels, and 48 channels of output are divided into 2 24. For the output of the 2 24 channels, the first 24 channels with the input of the first 16 channel full convolution, the second 24 channel and the input of the second 16 channel full volume. In extreme cases, the number of input and output channels is the same, for example, for the 24,group size is also 24, then each output volume kernel, only with the input of the corresponding channel convolution.

Caffe Official Website Quote:
Group (g) [Default 1]: If g > 1, we restrict the connectivity of each filter to a subset of the input. Specifically, the input and output channels are separated into G groups, and the i-th output group channels'll be only C onnected to the i-th input group channels.