Where else can you improve your dynamic group convolution?

[longmalongma]

Where else can you improve your dynamic group convolution?

[hellozhuo]

Where else can you improve your dynamic group convolution?

Hi thanks for your question, I am thinking how to make the DGC be more dynamic, i.e., dynamically allocate different amount of computation resources according to the difficulties of input images, which could be done by pruning different number of channels adaptive to the input. Now, the allocated computation resource is the same for all images.

[longmalongma]

dynamically allocate different amount of computation resources according to the difficulties of input images, which could be done by pruning different number of channels adaptive to the input.

Thanks for your reply, do you mean you want to prune the out channels adaptively?

[hellozhuo]

dynamically allocate different amount of computation resources according to the difficulties of input images, which could be done by pruning different number of channels adaptive to the input.

Thanks for your reply, do you mean you want to prune the out channels adaptively?

I think pruning either input channels or output channels adaptively could be fine.

[longmalongma]

dynamically allocate different amount of computation resources according to the difficulties of input images, which could be done by pruning different number of channels adaptive to the input.

Thanks for your reply, do you mean you want to prune the out channels adaptively?

I think pruning either input channels or output channels adaptively could be fine.

Thanks！