algorithms should take calibration into account

[maierl]

all algorithms, e.g. local thresholder should take the calibration of images into account. currently, the local threshold on xyz would typically use a much too big z-range compared to xy.

[dietzc]

another idea would be to upsample the image.

[maierl]

Which would increase the memory usage around 5-fold, especially with confocal images.

If this request is asking for too much, maybe at least the Z-dimension should be treated differently from X/Y, e.g. with the local threshold environment being 3 planes in Z and 50 pixels in X/Y.

[dietzc]

Virtual up-sampling won't increase the memory consumption at all. If you would save it it would precisely increase the memory consumption by the calibration of z.

However, if we implement algorithms with respect to calibration, we would treat the Z-Dimension relative to X,Y. I don't see another use-case... yet.

[maierl]

Is KNIME already using virtual up-sampling? Sorry if I didn’t notice.

What if the Z calibration is not divisible by the X/Y calibration? Wouldn’t that introduce artifacts?

[dietzc]

we don't. but we could (and should) :+1:

[maierl]

Virtual up-sampling sounds like a pretty advanced feature to me. Wouldn’t allowing for different x/y/z ranges be much easier / quicker to implement?