search

fiji / microscope-image-quality

TensorFlow-based classifier for microscope image focus quality.
https://imagej.net/Microscope_Focus_Quality
Apache License 2.0
17 stars 2 forks source link

Partial image analysis #5

Closed ctrueden closed 6 years ago

ctrueden commented 6 years ago

This branch adds the ability to analyze only a subset of the image. It works by allowing the user to choose a lesser amount of tile coverage in the X and/or Y directions.

spaced-tiles tiles-dialog

In the resultant probability image, values outside of analyzed tile boundaries are set to NaN.

This change breaks backward compatibility: there will no longer be a way to execute the analysis on the "top left" portion of an image, but only on a region with evenly spaced tiles. Here is an example demonstrating the discrepancy:

old new diff

@samueljyang @marktsuchida What do you think?

  1. Do you like this way of distributing tiles to resolve #4?
  2. Should we do something to preserve backwards compatibility? For example, we could also add explicit X and Y spacing parameters. But this would make the dialog more complex.
samueljyang commented 6 years ago

Yes, this looks great. I think it would be fine to break backwards compatibility.

ctrueden commented 6 years ago

Great. Would you have time to give it a quick spin with a dataset or two? If not I can merge as is, but always good to have another pair of hands try it first.

samueljyang commented 6 years ago

I just tested it, and it works great. I tested overlays, borders, and different numbers of tiles (I see it correctly gives an error if I set it to have more tiles than there is space in the image for). Please go ahead with the merge.