["Projection" terminology] MobileNet v2 Pointwise convulsion

In relation to the video on “MobileNet Architecture”, regarding the Mobile Net v2 network…

I understand the term “projection” is used to describe the final pointwise step (pre residual connection) but what is occurring is the opposite of projection and the more appropriate term would be “introjection”.

If you think of an image projector, it is usually increasing the size of the image that is displayed on a distant screen for a wider audience to observe, but here we are condensing the size of the image (or at least the output) of the prior depthwise convolution step.

Hence I believe “introjection” would be the more appropriate term. Think of it as the final pointwise convolution internalising the beliefs (output) of the prior depthwise convolution.

Just my thoughts :slight_smile:

Thanks for your thoughts.