Looking at the log file for Patch Motion, in the “rigid motion correction” section, there is a plot with 3 series - one for raw shifts, one for shifts with coarse smoothing, and one for shifts with fine smoothing.
The shifts with fine smoothing look like a consistently worse fit to the raw data than the “coarse” smoothing. None of these curves seem to change when I alter the “calibrated smoothing parameter”.
What is the difference between these two smoothing parameters, and is this something worth tweaking parameters to fix?
Hi @olibclarke, thanks for asking - how did you determine that the “fine” fit is worse?
The coarse and fine in this case refer to the resolution at which motion estimation is done - we first do “coarse” fitting at low resolution to help get large motions correct (especially important when there is a lot of motion or tilted data etc) and then do fine fitting after with higher resolution signal.
Can you try to adjust the “maximum alignment resolution” and see how that changes the results?
Thanks for the repy @apunjani! Well - I don’t know that it is worse, but it is certainly systematically different from the “coarse” fit, and for some of the mics really dramatically different. That makes sense re using the lower res data. I haven’t tried altering the alignment res, but certainly increasing the B-factor to 2000 makes the “fine” fit more similar to the coarse one (and actually doing that the CTF fits estimated by Patch CTF were very slightly better).
All these params only apply to the initial rigid whole frame fit, right? not the subsequent patch-based fit?
HI @olibclarke and @apunjani, this discussion seems to be very interesting to me. I want to understand how does the “max alignment resolution” and increasing the B factor helps to improve motion correction in patch CTF. What I understand is if you increase the max alignment resolution or B factor, the patches will be blurrier while aligning them and that way we can align them better. However, this will also align more of noise because of lower resolution. I was trying to play with these values with one of my datasets and I found that though increasing these values helps to align coarse and fine fit much better than default values, my CTF estimation was screwed up and I was observing that CTF fit curves were forced to align to a higher resolution value. The max resolution I used was 10A and B factor I tried 1000 and 2000. I also tried combining both. Would it be correct to say that if we have extreme amount of motion, in that case increasing either of these values would give better motion correction and better CTF fit subsequently? Or please let me know in what kind of situation we should consider altering these values?
