This was very well done and comprehensive. I liked the analysis of when it works well and when it doesn't, and the testing with nonsense images. We need more of

This was very well done and comprehensive. I liked the analysis of when it works well and when it doesn't, and the testing with nonsense images. We need more of this kind of rigor in ML.