Lanczos range

[axtimwalde]

Does that fix #212 ?

[ctrueden]

Without this patch, the following Groovy script works in my up-to-date version of Fiji (with imglib2 5.3.0):

#@ ImageJ ij

import net.imagej.Dataset
import net.imagej.ImageJ
import net.imglib2.interpolation.randomaccess.LanczosInterpolatorFactory

image = ij.scifio().datasetIO().open("/Users/curtis/data/clown.jpg")
ij.ui().show(image)

interp = new LanczosInterpolatorFactory()

double[] scaleDown = [0.5, 0.5, 1]
Object small = ij.op().run("scaleView", image, scaleDown, interp)
ij.ui().show(small)

double[] scaleUp = [2.5, 2.5, 1]
Object large = ij.op().run("scaleView", image, scaleUp, interp)
ij.ui().show(large)

double[] scaleWeird = [2.2, 0.6, 1]
Object stretched = ij.op().run("scaleView", image, scaleWeird, interp)
ij.ui().show(stretched)

And with the patch, it also works.

[tpietzsch]

Maybe the difference is that clown.jpg is not FloatType?

[ctrueden]

This script also works:

#@ ImageJ ij

import net.imagej.Dataset
import net.imagej.ImageJ
import net.imglib2.interpolation.randomaccess.LanczosInterpolatorFactory

image = ij.scifio().datasetIO().open("/Users/curtis/data/clown.jpg")
image32 = ij.op().convert().float32(image)
ij.ui().show(image32)

interp = new LanczosInterpolatorFactory()

double[] scaleDown = [0.5, 0.5, 1]
Object small = ij.op().run("scaleView", image32, scaleDown, interp)
ij.ui().show(small)

double[] scaleUp = [2.5, 2.5, 1]
Object large = ij.op().run("scaleView", image32, scaleUp, interp)
ij.ui().show(large)

double[] scaleWeird = [2.2, 0.6, 1]
Object stretched = ij.op().run("scaleView", image32, scaleWeird, interp)
ij.ui().show(stretched)

[ctrueden]

I rebased the lanczos-range branch over the latest master branch, and tested it with a generalized version of @tpietzsch's original non-working code from #212. Here is the code I used:

final float[] f = {10, 20, 30, 40, 50, 60, 70, 80, 90, 100};
final Img<FloatType> img = ArrayImgs.floats(f, f.length);
final RealRandomAccess<FloatType> a = Views.interpolate(Views.extendBorder(
	img), new LanczosInterpolatorFactory<>()).realRandomAccess();
for (int i=0; i<img.dimension(0); i++) {
	a.setPosition(new int[] { i });
	System.out.println(a.get());
	a.setPosition(new float[] { i + 0.5f });
	System.out.println(a.get());
}

I found that changing + 2 to + 3 for the lookup table array was insufficient to fix the problem. Empirically, for 1D data ranging in length between 4 elements and 20 elements, the minimum value that fixes the exception is + 22. Anything less than that still fails.

@axtimwalde Does that shed any light on the issue?