Could the padding factor be exposed for all the job types? Sometimes it’s very hard to get in memory with (I assume) the 2x padding default. Using 1.5x is a great compromise in such cases, but it seems like it’s only available in 2D classification.
just re-upping this… not being able to change the padding is a real pain for large box sizes, increasingly becoming an issue with higher res data becoming more frequent
Oli
this would still be great to have, it is a major limitation for large particles at high resolution
Hi @olibclarke thanks for bumping this - our underlying implementation (for all the “new” refinement job types) does support arbitrary zeropadding, we are working through connecting and exposing that functionality in all the job types
2 Likes
Hi @apunjani,
I am doing NU Refinement (NEW) for my dataset and noticed that there is no padding during the refinement:
[CPU: 1.70 GB] ====== Refinement ======
[CPU: 1.70 GB] Input particles have box size 720
[CPU: 1.70 GB] Input particles have pixel size 1.2200
[CPU: 1.70 GB] Particles will be zeropadded/truncated to size 720 during alignment
[CPU: 1.70 GB] Volume refinement will be done with effective box size 720
[CPU: 1.70 GB] Volume refinement will be done with pixel size 1.2200
[CPU: 1.70 GB] Particles will be zeropadded/truncated to size 720 during backprojection
[CPU: 1.70 GB] Particles will be backprojected with box size 720
[CPU: 1.70 GB] Volume will be internally cropped and stored with box size 720
[CPU: 1.70 GB] Volume will be interpolated with box size 720 (zeropadding factor 1.00)
Would it be better to have some padding (1.5x at least) for the refinement? Thanks.
Sorry. I might have misunderstood this: Zeropadding does not mean no padding.