What about changing the depth of a layer (number of channels)?
It seems that this isn’t covered by d, w and r?
Welcome to the community.
It’s good point. I think Andrew’s picture is slightly misleading.
“Width” is the number of channels.
The paper for Efficient Net refers this paper for Wide Residual Networks.
If we look at (a) basic and (c) basic-wide, we can see the output from conv3x3 is wider in (c). This is channels.
Thank you for your full and timely response.
It helped me to understand the topic.