Thanks for sharing!

Hi panovr!
First numbers are image shape as usual. And the last one quantity of the features generated by 1x1 convolution. In paper on page #4, paragraph “bottleneck layers” was mentioned that “Unless otherwise specified, each 1×1 convolution reduces the input to 4*k feature-maps in all experiments”. So because for my example I’ve take feature map k=12, I’ve write features generated from 1x1 convolution layer as 4*12.

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.