5:19, you were up early, @haesleinhuepf! Such devotionā¦
Unfortunately the Ext.CLIJ2_clear() statement right after initialization of the CLIJ2 extensions doesnāt help. I still get
net.haesleinhuepf.clij.clearcl.exceptions.OpenCLException: OpenCL error: -36 -> CL_INVALID_COMMAND_QUEUE, sometimes followed by
net.haesleinhuepf.clij.clearcl.exceptions.OpenCLException: OpenCL error: -5 -> CL_OUT_OF_RESOURCES..
I actually forgot to tell you that Ext.CLIJ2_mean3DBox works well, but it crashes on Ext.CLIJ2_median3DBox (or anyother CLIJ2 median command).
And you are right, the downsampling I added only later. (By the way, the console also says downsample3D is deprecated? It doesnāt look like that in the docs). Here is the complete code, without pushing image1 twice:
run("CLIJ2 Macro Extensions", "cl_device=");
Ext.CLIJ2_clear();
run("T1 Head (2.4M, 16-bits)");
image1 = getTitle();
Ext.CLIJ2_push(image1);
radius = 2;
// compute 3D mean
Ext.CLIJ2_mean3DBox(image1, image2, radius, radius, radius);
Ext.CLIJ2_pull(image2);
// compute 3D median
Ext.CLIJ2_downsample3D(image1, image3, 0.5, 0.5, 0.5);
Ext.CLIJ2_median3DBox(image3, image4, radius, radius, radius);
Ext.CLIJ2_pull(image4);
Ext.CLIJ2_clear();
Which automatically brings me to another point: In analysis workflows it often occurs that you want to execute a statement conditionally, e.g. background subtract, filter, etc, based on the userās input.
For this particular macro, if you want to run it with and without downsampling, besides outcommenting the downsampling you also need to rename the input image in Ext.CLIJ2_median3DBox to image2, because image3 doesnāt exist. Therefore, one cannot easily have a boolean user input downsampling and a statement if(downsampling == true) Ext.CLIJ2_downsample3D(...).
I guess this is a consequence of the (otherwise great) ābyRef Imageā method. In the past you had made some CLIJ2 commands āin placeā, but these seem to have been phased out.
Now I was thinking, would it in principle be possible for CL(IJ) to execute commands āinplaceā if the destination image variable is set the same as the input image? (e.g. by recognizing that, clearing the input image memory and then renaming the output reference to the input reference.) Or is that inherently very difficult āunder the CL hoodā? It would definitely solve many issues.
Thanks again!
Bram
