Add converters between ImgLib2 RAI and ImageJ1 ImagePlus

[ctrueden]

There are converters for Dataset to/from ImagePlus (though see #186). But if you have e.g. an ArrayImg, you cannot currently convert it directly to ImagePlus with the ConvertService. We want this to be possible.

[imagejan]

I guess this can be done with the new AbstractDelegateConverter introduced in scijava/scijava-common#366, right?

Which conversions need to be supported explicitly?

• ArrayImg => ImagePlus
• PlanarImg => ImagePlus
• CellImg => ImagePlus
• ...?

[ctrueden]

I think the delegate converter will be useful, yeah. Thanks for developing it.

For new converters needed:

1. ImgPlus to ImagePlus – to map the metadata e.g. axes correctly.
2. RandomAccessibleInterval to ImagePlus – for all other cases i.e. no metadata is available.
3. ImagePlus to ImgPlus. We already have ImagePlus to Dataset, which extends Img, so if you request a conversion of an ImagePlus object to e.g. RandomAccessibleInterval, it should already work (though I didn't test just now). But because Dataset does not extend ImgPlus (inorite), that request would currently fail.

The above is just speculative—best would be to do some thorough TDD requesting all sorts of anticipated desirable conversions, and then implement the most general converters that also minimize information loss.

[imagejan]

For the record, this Groovy script lists current support for some of the conversions:

#@ ConvertService cs

import ij.ImagePlus
import net.imagej.Dataset
import net.imagej.ImgPlus
import net.imglib2.img.Img
import net.imglib2.RandomAccessibleInterval

println cs.supports(Dataset.class, ImagePlus.class)                  // true
println cs.supports(Dataset.class, ImgPlus.class)                    // true
println cs.supports(Dataset.class, Img.class)                        // true
println cs.supports(Dataset.class, RandomAccessibleInterval.class)   // true
println ""
println cs.supports(ImagePlus.class, Dataset.class)                  // true
println cs.supports(ImagePlus.class, ImgPlus.class)                  // false
println cs.supports(ImagePlus.class, Img.class)                      // true
println cs.supports(ImagePlus.class, RandomAccessibleInterval.class) // true
println ""
println cs.supports(ImgPlus.class, Dataset.class)                    // false
println cs.supports(ImgPlus.class, ImagePlus.class)                  // false
println cs.supports(ImgPlus.class, Img.class)                        // true
println cs.supports(ImgPlus.class, RandomAccessibleInterval.class)   // true
println ""
println cs.supports(Img.class, Dataset.class)                        // false
println cs.supports(Img.class, ImagePlus.class)                      // false
println cs.supports(Img.class, ImgPlus.class)                        // true
println cs.supports(Img.class, RandomAccessibleInterval.class)       // true
println ""
println cs.supports(RandomAccessibleInterval.class, Dataset.class)   // false
println cs.supports(RandomAccessibleInterval.class, ImagePlus.class) // false
println cs.supports(RandomAccessibleInterval.class, ImgPlus.class)   // false
println cs.supports(RandomAccessibleInterval.class, Img.class)       // false

[imagejan]

I started implementing a parameterized test in fb75ae4, where different types and axis orders can be defined to be tested in ImgPlus => ImagePlus conversion.

[ctrueden]

See also imagej/imagej-common#74

[ctrueden]

And imagej/imagej-ops#54

[imagesc-bot]

This issue has been mentioned on Image.sc Forum. There might be relevant details there:

https://forum.image.sc/t/net-imagej-defaultdataset/37085/3

[imagesc-bot]

This issue has been mentioned on Image.sc Forum. There might be relevant details there:

https://forum.image.sc/t/save-ilastik-hdf5-files-from-java/60194/2

[ctrueden]

I wrote a more thorough script printing supported conversions:

image-conversions.py

#@ ConvertService cs

import sys
from ij import ImagePlus
from net.imagej import Dataset, ImgPlus
from net.imagej.display import DatasetView, ImageDisplay
from net.imglib2.img import Img
from net.imglib2 import RandomAccessibleInterval

image_types = [
	ImagePlus,
	RandomAccessibleInterval,
	Img,
	ImgPlus,
	Dataset,
	DatasetView,
	ImageDisplay,
]
sys.stdout.write("              ")
for col in image_types:
	sys.stdout.write("{0: <13}".format('RAI' if col == RandomAccessibleInterval else col.__name__))
print
for row in image_types:
	sys.stdout.write("{0: <14}".format('RAI' if row == RandomAccessibleInterval else row.__name__))
	for col in image_types:
		sys.stdout.write("yes          " if cs.supports(col, row) else "NO           ")
	print

And the results as of this writing:

              ImagePlus    RAI          Img          ImgPlus      Dataset      DatasetView  ImageDisplay 
ImagePlus     yes          NO           NO           NO           yes          NO           yes          
RAI           yes          yes          yes          yes          yes          NO           NO           
Img           yes          NO           yes          yes          yes          NO           NO           
ImgPlus       NO           NO           yes          yes          yes          NO           NO           
Dataset       yes          NO           NO           NO           yes          NO           NO           
DatasetView   NO           NO           NO           NO           NO           yes          NO           
ImageDisplay  yes          NO           NO           NO           NO           NO           yes        

The PyImageJ and napari-imagej projects have to bend over backwards to support image conversions that aren't supported by the ConvertService yet, so I'm motivated to make the above grid all "yes".

[ctrueden]

imagej/imagej-common#102 adds new converters for Dataset to DatasetView and RAI to ImageDisplay. Updated table:

              ImagePlus    RAI          Img          ImgPlus      Dataset      DatasetView  ImageDisplay 
ImagePlus     yes          NO           NO           NO           yes          NO           yes          
RAI           yes          yes          yes          yes          yes          NO           NO           
Img           yes          NO           yes          yes          yes          NO           NO           
ImgPlus       NO           NO           yes          yes          yes          NO           NO           
Dataset       yes          NO           NO           NO           yes          NO           NO           
DatasetView   NO           NO           NO           NO           yes          yes          NO           
ImageDisplay  yes          yes          yes          yes          yes          NO           yes          

[ctrueden]

With imagej/imagej-common#103, and specifically imagej/imagej-common@e67b5c9, I have now coded converters for all image types to/from all others, except for ij.ImagePlus since that's not available from imagej-common. But it gets us much closer to the above table becoming all yeses!

[imagesc-bot]

This issue has been mentioned on Image.sc Forum. There might be relevant details there:

https://forum.image.sc/t/converting-arrayimg-to-imageplus-for-using-the-analyze-skeleton-plugin/85934/3