Since I've coded this implementation, official code has been released. This also features a more efficient calculation and more accurate results. Therefore, I recommend the official implementation
Implementation of paper "Do Wide and Deep Networks Learn the Same Things? Uncovering How Neural Network Representations Vary with Width and Depth" by Thao Nguyen and Maithra Raghu and Simon Kornblith. Please find my post explaining the implementation on Medium/TDS.
The current implementation focuses on creating figures similar to those in the paper:
