Bead Calibration pixel Size

[pranayj77]

Does the pixel size for the calibration beads have to be 100*100nm.
In the random simulation file is seems like the voxel size is hardcoded to be [1,1,self.dx]

[Haydnspass]

Dear @pranayj77
the pixel size can be set in the parameter file, for reference see here:

DECODE/decode/utils/reference_files/param_friendly.yaml

Line 6 in 8977766

px_size:

[pranayj77]

I tried changing that value but it will not change the size of the psf.


The above images were generated with the pixel size in the param file set to 100nm100nm and 6565nm respectively

[pranayj77]

The voxel size here seems to be hard coded as (1.,1.,self.dz).
Does this affect the psf generation from the Cspline fitting?

[Haydnspass]

This is just internal conversion within the spline function. Since the calibration stemming from SMAP are in px units, the x and y voxel "size" is set to 1.
For training data simulation we sample directly in px units.
Only after the model, i.e. post-processing voxel size comes into play.

[Haydnspass]

I was dropping @jries a note because he is the better expert for calibrations and stuff.
But I suspect something else is going wrong here, hm.
Have you checked the validation of the psd in SMAP? There should be an output after the cubic spline is calibrated

[pranayj77]

I have checked the validation in SMAP. It looks ok to me based on the the relatively small number of frames we provide to it, since our data is 2D. How would I check if something is significantly wrong in the validation?

[jries]

Dear all, the PSF calibration plugin in SMAP is agnostic of the pixel size. It works with pixels as a unit. The PSF is stored in units of pixels. DECODE uses the same calibration and does the simulation also in units of pixels. Therefore, the entire PSF generation and training process should be independent of any pixel size you set, and the simulated and real images should look similar without any rescaling. If not, check if the PSF generation worked (in the SMAP plugin you can explore the PSF and also check if the average PSF represents the individual beads). And if during the simulation the z-coordinate is correct (maybe you compare simulated and experimental images at different z-positions).

[Haydnspass]

@pranayj77 any updates on this?

[pranayj77]

I am so sorry I completely forgot to reply.
My PI decided that Decode wasn't offering significant improvements over our current method on 2D images so she told me to abandon the project.