Usability of jai-imageio-jpeg2000?

[ctrueden]

Apologies if we've discussed this before (if so, it's been long enough that I've forgotten).

• The project https://github.com/jai-imageio (jai-imageio-core and jai-imageio-jpeg2000) is the closest thing we have to an official place for the JAI ImageIO codebase.
• I notice this repository is described as "OME Java ImageIO fork (deprecated)". What is the plan for moving away from this component?
• Would it make sense to switch OME dependencies to com.github.jai-imageio:jai-imageio-core and com.github.jai-imageio:jai-imageio-jpeg2000 rather than continuing to maintain this deprecated fork?
• For what it's worth, the SCIFIO project made the switch in early 2024 with scifio/scifio@d5dfe9a. There were very small (<1e-5) differences in JPEG2000 compression test results.
• Looking at the jai-imageio people, I notice that @sbesson, @jburel and @mtbc are co-owners already, and no one has pushed any changes to the jai-imageio repositories in ~5 years, so we should be free to push any needed fixes to complete the switch, no?
• Are there any licensing/patent reasons that could impede continuing use of the ImageIO/JPEG2000 codebase?

Edit: it looks like this fork of JJ2000 has significantly improved the code... has anyone tried it?

[ctrueden]

Also of potential interest:

• https://github.com/geosolutions-it/imageio-ext
• https://github.com/uclouvain/openjpeg (and de.digitalcollections.imageio:imageio-openjpeg)

[melissalinkert]

We had previously tried switching to https://github.com/jai-imageio/jai-imageio-jpeg2000 in ome/ome-codecs#21 and ome/bioformats#3857, but it wasn't completely clear from the comments why that wasn't completed.

I tried again with https://github.com/ome/ome-codecs/compare/master...melissalinkert:ome-codecs:jai-try2?expand=1 and https://github.com/ome/bioformats/compare/develop...melissalinkert:bioformats:jai-try2?expand=1.

mvn clean package install on the ome-codecs branch works, and installs a 1.1.2-SNAPSHOT. After updating the ome-codecs version in the bioformats branch to 1.1.2-SNAPSHOT, mvn clean package fails this roundtrip lossless JPEG-2000 compression test:

https://github.com/ome/bioformats/blob/develop/components/formats-bsd/test/loci/formats/utests/LosslessJPEG2000Test.java

with mismatching pixel values. A few subtle differences in lossy behavior might be fine, but I don't see being able to adopt https://github.com/jai-imageio/jai-imageio-jpeg2000 as-is when there is a known case of lossless compression/decompression not behaving correctly.

I haven't had time to try https://github.com/faceless2/jpeg2000 yet; an immediate concern there is it doesn't appear to be deployed to anywhere (at least that I could find).

[ctrueden]

@melissalinkert Thanks for trying! If you like, I could try diffing ome-jai with jai-imageio-jpeg2000 to find the differences? If I had to guess, the source of the differences might just be bug-fixes you did over the years to ome-jai that haven't made their way into jai-imageio-jpeg2000. What's the best file for me to use for testing?