Hi, Thanks for your reply.
However, I think there is a misunderstanding. I apologize that my question was not clear enough.
My question means that I want to apply different kernels to each image, not different channels.
More specifically, the arguments of the parallel_conv2d are:
inputs: mini_batch x in_channel x iH x iW tensor - batch of input “images”
filters: mini_batch x out_channel x in_channel x KH x kW tensor - a batch of convolutional kernels
What I want to do is applying filters[i] to inputs[i].