How does max pooling increase the number of channel in the feature map?

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

I was re-watching the U-Net architecture video in week 3. At around 3:05 time Prof. NG said that pooling will increase the number of channels. But from What I saw on other topics posted here e.g. “Channels depth increment after pooling” pooling doesn’t increase the number of channels. So why does Prof. Ng say in the video that pooling increased the number of channels? Is it a mistake?

2

Hi @Shivam_sharma_MED_07 ,

I went to the segment of video that you mention and I can see how it can lead to your question.

Max Pooling cannot increase the number of channels. Max Pooling, as you learned, will only reduce the dimension of its input.

The U-Net, as seen in the lab of that course, is comprised of a Downsampling followed by an Upsampling.

In the lab, the downsampling was built by blocks of:

Conv2D
Conv2D
(Dropout)
MaxPooling

So may be in the explanation, Prof Ng was referring to the effect of the Convolutions but didn’t explicity said it - I am just guessing.

But for sure, a Max Pooling would not increase channels but instead reduce the dimensions of the input.

Thoughts?

Juan

3

Thank you very much for your answer. Now it is more clear to me.

1 Like