Checking for time points with sufficient enough deepnet confidence

For the tracking nnet output, you shouldn’t be using timepoints with low confidence/probability. I’d make sure to only use time points with all 3 points p>0.9 (try looking at the distribution to see what'd be a reasonable threashold)

Locate the tracking neural-network output and inspect the confidence distribution for the three tracked points. Define a reasonable threshold from that distribution, then verify that only timepoints where all three probabilities exceed it are used.

Written by the indexing model from the issue text.