What is channel-wise feature interaction in CNNs and how does it evolve in deep networks?

Channel-wise interaction refers to how CNNs combine features across channels to form richer representations.

Each convolution layer outputs multiple feature channels. Early layers detect simple patterns per channel, but deeper layers combine these channels to form complex representations. 1x1 convolutions and pointwise operations are key mechanisms enabling cross-channel mixing. Without channel interaction, each feature map would remain independent and less expressive.