EfficientNet - Notation

[jackchan.hk]

Just curious why Andrew denote the height of a convolution block as w? Should it be h instead?

[anon57530071]

Andrew’s image is slightly misleading.

Height and Width for an image is part of “r”, resolution.
Depth “d” is the “depth of network layers”.
Width “w” is the number of channels created by multiple filters in convolutions.

Here is the another thread about the width.

Hope this helps.

[jackchan.hk]

Dear Nobu Asai

Tones of thanks, its so much clear for the overall picture.

Regards,
Jack