Hello, my first day with PerceptiLaps (TwoMinutesPapers brought me here).
Works great, I could follow the videos, will spend more time with it.
One observation: The very first thing I wanted to see after MNIST training was the failed examples - to see if I could understand what a problem might be with any given item.
But the “Next sample” button took me only through good matches between ground truth and prediction - which is satisfying at first - but not productive for me to improve.
So is there / will there be a way to filter and show only the “bad apples” in some metric?
Thank you all, and keep up the great work!