Clij Clij Versions Save

Bugfixes:

• Several OpenCL kernels caused issues on the beignet 1.3.2 platform (Intel GPU driver) on Ubuntu 18.04
• Memory leak while context switching (closing and restarting CLIJ) for NVidia GPUs - thanks to Loic Royer for fixing!

Bugfixes:

• Affine transforms, scaling, rotation and translation didn't work for 2D images on NVidia GPUs. Thanks to Peter Haub for reporting and fixing the bug.
• Gaussian blur, mean and median filters delivered potentially wrong results when being applied to 8-bit images as they were interpreted as int8 images instead of unsigned int8. Thanks to Tanner Fadero for reporting this bug.
• Intermediate results of separable filters were saved in wrong image type resulting in small errors due to numerics when being applied to 8-bit or 16-bit images. Intermediate results are now saved as float minimizing these errors. Thanks to Martin Weigert for reporting this bug.
• Binary operations erosion and dilation may have delivered images with pixel values different from 0 and 1 in case non-binary images were given as input.
• Type casting was unified between different GPU hardware. We implemented the ImageJ-way of type casting (clamp to min/max): Adding a value of 1 to an pixel of type 8-bit with value 255 leads to value 255. This affects basically all OpenCL-kernels and especially non-Intel GPUs.
• Not needed content was removed and thus, clij_.jar and clij-core.jar decreased size on disc from about 15 MB to less than 1 MB.

API Changes:

• affineTransform() has been deprecated, use affineTransform2D() or affineTransform3D() instead.
• scale() has been deprecated, use scale2D() or scale3D() instead.
• clij.pullBinary() (Java/Jython/Groovy/...) and CLIJ_pullBinary() (macro) have been added to improve interoperability with ImageJ. Thanks to Nico Stuurman for suggesting this.

Backwards compatibility breaking API changes

• crossCorrelation() was removed from the Java/Jython/Groovy API for technical reasons. It has been moved to the clij-advanced-filters repository for revision.

Bugfixes:

• Several input and output images can be of different types now
• Support for big images (byte > 2 GB, short > 4 GB and float > 8 GB)
• Improved support for older GPU models (OpenCL 1.1)
• Improved error messages

Bug fixes:

• Support for big images (byte > 2 GB, short > 4 GB and float > 8 GB)
• Improved support for older GPU models (OpenCL 1.1)
• Improved error messages

Bugfixes:

• clij.push(), clij.pull() and clij.convert() were not working when running CLIJ from MicroManager

New features:

• clij.pushCurrentSlice() and CLIJ_pushCurrentSlice()
• clij.op().centerOfMass()

Bug fixes:

• Multiple GPUs didn't work as expected because of a static context and some synchronisation issue

Removed methods:

• particleImageVelocimetry was removed. Current development version lives in the clij-advanced-filters repository. It will come back.

Bugfixes:

• Pushing an image to GPU doesn't change C/Z/T sliders of the image anymore
• Kernel size of Gaussian blur increased. It is now int(4 times sigma)+1. Thanks to Martin Weigert for finding this bug.
• Menu entries have been renamed for clarity

New features

• Binary erosion and dilation slice by slice
• Add median filters for box neighborhood
• Add CLIJ version number to menu to allow easier debugging

New features:

• create2D and create3D for allocating memory in GPU
• clij.pullRAI() for ImageJ2 support
• Experimental auto-completion for groovy and jython

API methods have been renamed and removed, breaking backwards compatibility:

• blurFast() was removed, use blur() instead
• blur(img, img, int, int, int, float, float, float) was removed. use blur(img, img, float, float, float)
• blur(img, img, int, int, float, float) was removed. use blur(img, img, float, float)
• dilateBoxIJ renamed to dilateBox
• erodeBoxIJ renamed to erodeBox
• maximum2DIJ removed. Use maximum2D instead
• mean2DIJ removed. Use mean2DBox or mean2DSphere instead
• minimum2DIJ removed. Use minimum2D instead
• thresholdIJ renamed to threshold

Removed features:

• displacement (use particleImageVelocimetry instead)
• spotDisplacement (use particleImageVelocimetry instead)

Added features:

• centerOfMass
• particleImageVelocimetry