What is the main role of the convolutions placed before the and convolutions in the dimension-reduction Inception module?

Question

What is the main role of the    convolutions placed before the    and    convolutions in the dimension-reduction Inception module?

Answer

To reduce the number of channels and therefore lower the cost of the following convolutions.

Answer

To merge the branch outputs by element-wise addition.

Answer

To halve the spatial resolution of the feature maps.

Answer

To enlarge the spatial receptive field without changing the channel count.

Crowdly