Primary Capsules reconstruction

[Ellyuca]

Hi, I have a few question about the reconstruction of primary capsules. The idea is really interesting and I would like to understand
better the logic behind it.

You said the following:

"We used a pre-trained network to train a reconstruction module for Primary Capsules. By scaling these capsules by its routing coefficients to the classified object, we were able to visualize reconstructions from Primary Capsules. Each row is reconstructed from a single capsule, and the routing coefficient is increased from left to right."

Could you explain and give more details on how this was done?
Could you also point out the code section that does the primary capsules reconstruction? I looked over the code but I can't figure it out.

Thank you very much for your time.
I'm looking forward to reading your answer.