Squezee-and-excite networks introduced SE blocks, while MobileNet v2 introduced linear bottlenecks.

What is the effective difference between these two concepts?

Is it only implementation (depth-wise convolution, vs per-channel pooling), or they serve a different purpose?

My understanding is that both approaches are used as attention mechanism, working per-channel. In other words, both approaches are used to filter unnecessary information (information that we consider noise, not signal). Is this correct? Do bottlenecks ensure, that the same feature won't be represented multiple times in different channels, or they don't help at all in this regard?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.