.vmic WSI reader

[kwiechen]

Hello,

the .vmic WSI file (#4264) is a nested Zip archive including a deep zoom pyramid. .vmic format information is from:

openslide/openslide#168
https://lists.andrew.cmu.edu/pipermail/openslide-users/2015-December/001164.html

The VmicReader reads the deep zoom pyramid using:

https://github.com/usnistgov/pyramidio

Reading data from WSI files < 2 GB is possible directly using java.nio.file.FileSystems. The reader unzips the outer zip container into a temporary file if size is > 2 GB.

Example data: https://zenodo.org/records/14714280

Best regards,

Kai

[kasper-cancilico]

@sbesson @melissalinkert following up on Melissa's note connecting this to #4453. Kai and I have agreed to consolidate onto #4397 (mine adds a synthetic-fixture unit test and a few robustness fixes on top of his reader), and since it's milestoned for 9.0.0 I wanted to ask whether it could get a first review. That would tell us what to change before I open the follow-up against Kai's branch, and it avoids duplicating effort. No urgency on merging. Thanks!

[melissalinkert]

Thanks @kwiechen and @kasper-cancilico for reviving this discussion, and apologies for the delay.

Please see initial inline review comments; also please revert the move of SubResolutionExample.java.

In places where there is some unexpected/undesired behavior, I've tried to note commands you can run on your own to reproduce that behavior. Another test you can run is to take the files from https://zenodo.org/records/18241389, put them in a directory, and then run (from the root of this repo):

$ cd components/test-suite
$ ant -Dtestng.memory=2g -Dtestng.directory=/path/to/vmic/data -Dtestng.configDirectory=/path/to/vmic/data test-config
$ ant -Dtestng.memory=2g -Dtestng.directory=/path/to/vmic/data -Dtestng.configDirectory=/path/to/vmic/data test-automated

See https://bio-formats.readthedocs.io/en/latest/developers/commit-testing.html#automated-tests for details. That is what we would do to get data into our test repo, so a prerequisite for merging is that both of those ant ... commands run without errors.

Finally, if you are planning to add commits here, @kasper-cancilico, we would need a signed Contributor License Agreement (we already have one for @kwiechen).

[melissalinkert]

Since this is whole slide data, I'd expect FormatTools.HISTOLOGY_DOMAIN here.

[melissalinkert]

Here and in other imports above, we prefer to have necessary imports enumerated instead of using *.

[melissalinkert]

Please use the existing logging infrastructure instead of System.out.println.

[melissalinkert]

If close(false) is called, all fields should be reset to their original value, including e.g. filesFolder. I would expect that lines 145-158 should only happen if close(false) was called (and not close(true)), but if there is a reason that this is always done then it would be helpful to clarify.

[melissalinkert]

Overriding getResolutionCount() should be unnecessary.

I note that showinf -nopix -noflat vmic-sample-small-1.vmic shows a single pyramid of 6 resolutions. I would then expect showinf -nopix vmic-sample-small-1.vmic to show 6 separate series (one resolution per series), but instead I see a single series.

[melissalinkert]

As with the PixelPerMicron key, the value for Magnification could possibly be null.

[melissalinkert]

Is there a reason to re-implement ImageIO logic here, instead of making use of existing Bio-Formats readers that use ImageIO (e.g. JPEGReader)? That would also allow removal of most of the BufferedImage handling, since the existing ImageIO-based readers already unpack BufferedImage to a byte array.

[melissalinkert]

File variables will break memoization, which showinf -nopix -noflat -cache /path/to/file.vmic can demonstrate. For filesFolder, it should be enough to remove this line, and replace use of filesFolder with DZI_FILES.

[melissalinkert]

Similar to filesFolder above, the File variable is a problem for memoization. If innerZipFile is needed outside of initFile, you could store the String path instead.

[melissalinkert]

This will also cause a problem for memoization, but is a little less straightforward to fix. The most consistent with other readers would be to make this field transient, then override the reopenFile() method to set inner_zipfs appropriately. ZipReader and MinimalTiffReader are two examples that follow a similar pattern.