The junk detector is very useful, but is still quite slow on large datasets.
Would it be possible to run it on thumbnails generated by patch motion, if available, and then just scale the masks to match the micrographs? The junk features are generally pretty low res so I’m not sure you’d lose much fidelity wise
Oli
Hi @olibclarke,
Unfortunately the current junk detector is trained at a fixed pixel size, so we can’t do this without retraining. But, how slow is slow? (and how big is the dataset)? We haven’t found the junk detector to be unduly slow - I just want to make sure nothing pathological is going on…
–Harris
Not that slow (hours, not days) for ~20k mics - I guess I was just being impatient, disregard! 