Hoew 1x1 filters reduce the depth of final inception layer in Googlenet?

It was told that a final 1x1 convolution with D0 filters will be applied on the aggregated sum of layers after different convolution operations to reduce the final D1 layers depth to D0 layers.

But actually the no. of layers are increasing after each operation.

How 1x1 convolutions reducing the depth in GoogleNet CNN?

#The intuition behind GoogleNet

1x1 convs are used in a way that input and output dimensions remain same.
So if we use 1x1xD0 such filters, it can be achieved.

Thank you, Ishvinder.
But the video lecture (The intuition behind GoogleNet) is not saying the same.
Please share any useful article on this topic.

Hi @Sai_Deepak,
Maybe this can help you out.