It’s been a while but I have an update I hope will help you track this down.
We hit the issue again last week, but this time I found a specific movie that I suspect is the cause. This (very poor quality) movie evaluates to an ice thickness of NaN when checking the stats from the Live interface:
The CTF plot looks like this:
which doesn’t look great for estimating a ratio - divide by zero error?
So I guess the reason we previously only saw this after the session had been running for a while was just that it needed a bad enough movie to enter the processing.
Another problem caused by this movie is that the motion estimates are extreme:
Total Motion (pix): 1.4891705695163038e+36
Max In-Frame Motion: 9.596886779369273e+34
causing the scale in Overview for those parameters to become useless, as it would take “forever” to zoom in to see the rest of the data:
Having some sanity check or reasonable max values on some of the parameters would be nice.
Along the same line, it would be nice with a bit more freedom in setting the thresholds in the Overview. For now, it seems to only allow values close to the values observed so far in the session, which often mean we can’t set good cut-offs until we see the first outliers in a running data collection.
Even better would be to have default values as part of our presets, so we could apply conservative cut-offs from the beginning (eg. defocus range 0-2000, total motion 0-40, astigmatism 0-1000, ice thickness 1.00-1.12 ect.).
Sorry for going a bit off-topic and into suggestions for usability improvements instead!