Some questions about OpenVX

Hello, I’m Sanghyun Park, from Korea.

I have several questions regarding using OpenVX.
As I understand, Khronos is ready to release/distribute the header files.
And in the header files, the data objects are already defined, and what I have to do is
to implement the vision algorithm obeying the kernel interface you’ve provided.
(and also to implement the vision application by construct a graph using the kernel nodes).
Is this correct?

Well, I’m planning to port OpenVX to some DSPs. In order to make a precise schedule,

  1. could you let me know when the header files will be available?
  2. also when the “sample” implementation of the kernels will be available?

I found that there are about 40 kernels that should be implemented, and I think it should be
extended a lot to have more kernels. I should’ve suggested some APIs that I need, but I’m focusing on how to
boost the performance up using OpenVX on our DSPs, not on finding the missing kernels.
So, sorry for not giving you some feedbacks you are expecting.

My final question is,
3. Currently, most of the APIs are implemented to support U8 data type in the current release (1.0).
Will it be extended to support all the data types in the next release?

I have many questions and plans about OpenVX.
It would be a good starting thread to begin with.

Thanks,

Best regards,

1.) The headers will be available after final ratification. We are still taking comments and changing the API, so releasing headers now would cause rework.
2.) The Open Source Sample OpenVX will be released around final ratification as well.

We do need more kernels, but we have to be careful to only accept kernels which can be accelerated by vendor hardware & platforms. That is the purpose of the API. We DO want your suggestions for new kernels! If a kernel can’t be accelerated, it may still be useful, but could be supplied as an (Open) 3rd party “C” implementation which can be loaded using “vxLoadKernels”.

3.) There will probably be an extension for support for signed 16 bit on most functions in the near future.