Let me introduce PharoVX: a Pharo binding to OpenVX, the standard for accelerated computer vision.
OpenVX is currently [being] implemented by various vendors, but right now we can download a sample implementation from Khronos for free and use it in Pharo.
PharoVX is mostly written during ESUG 2015 in Italy, and I just wrapped it up a bit on this week before posting to list.
The project homepage provides some instructions on how to obtain and build the reference implementation, how to do the basic setup, and how to debug your VX pipelines.
PharoVX now also allows to run all kernels specified in standard v1.0.1.
There are some known problems:
1. Mapping data to-and-from VX is slow – it is currently done pixel-by-pixel in a Smalltalk code. I will probably try to speed it up with NativeBoost.
2. On the first invocation, the image data can be corrupted. For the subsequent VX executions, it seems to be fine.
With best regards,