Share the code to reproduce the different figues of the paper

[zippeurfou]

Hi,
One of the thing I really liked in the paper is the different figures.
From figure 1b that allow to explore the DCNv2 layer, figure 2 that provide IA for each fields, to figure 4 for example that shows IA through training.
I am asking for code because I think this could be really useful for model debugging and would be a great tensorboard visualisation as well.
Thanks for the paper, I really enjoyed reading it. Lot of great finding.