Ghosting caused by blending

[joshua-gould]

Hi, ashlar succeeds in aligning tiles, but the resulting stitched image creates ghost nuclei:

You can see the two original tiles here:

Besides trying different values of --filter-sigma and --maximum-shift, are there other approaches we can try to eliminate these artifacts? Does ashlar compute any values that indicates the stitching quality that we can use or do we have to manually inspect the stitched image? Thanks.

[jmuhlich]

I think your stitching actually failed and the ghosting is from wrongly-aligned cells, not the blending itself. It's hard to recognize misalignment when the cells are this sparse and the background is very dark. We're working on a more formal QC report that will unmistakably notify you when this sort of thing happens, but it's not quite ready.

If you would be willing to share your images and the ashlar command line you used, I can try to figure out what went wrong. Let me know and we can coordinate on how to transfer the images.

Alternatively, if you're comfortable installing an experimental version of Ashlar you can try out the QC report feature and share the report here.

[joshua-gould]

I'd be happy to try the experimental version. Are the stitching results from the experimental and released version the same? Thanks.

[joshua-gould]

@jmuhlich Can you please point me to the experimental version of Ashlar to test? Thanks.

[jmuhlich]

You can install it with this command:

pip install "ashlar@git+https://github.com/jmuhlich/ashlar.git@qc-report"

Then add --report ashlar-report.pdf to your ashlar command line. Upload the resulting ashlar-report.pdf here. The stitching results will be the same, or perhaps somewhat improved depending on which version you were using before.