This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

TDA4VM: How to build camera image processing pipeline in GPU?

Part Number: TDA4VM

Hi Sir/Mamdam,

I'm planing to use GPU(GE8340) in TDA4VM SoC to do some image processing(Ex; Local Tone Mapping), can you please provide a example:

conneted from camera image stream, build a camera image processing(Ex: Gaussion Filter) pipeline in GPU

Thanks,

Chris

  • Hi Chris,

    Can you share the TI release baseline you are using.

    The normal way would be through use of open gles model. The GPU in TDA4VM supports OpenGLES 3.0 API, and the typical procedure would be you would read the camera images as textures and do the relevant processing in combination of vertex and fragment shaders. Is there a specific processing that you are interested, TDA4VM has few additional accelerator blocks for certain capture processing that might also be worth looking at.

    Thanks,

    Gowtham

  • Hi Gowtham,

    Thank you for your so fast reply!

    I;m sorry TI release baseline can't be provided in now due to I'm on holidays, and I'll share it once I'm back to office.

    Actually I'd like to know how to build this pipeline, is it use OpenVX framework? How to config it? Any limitaions? Is GLSL availiable on QNX OS?

    The processing is: Local Tone Mapping, Temporal Noise Reduction and others. I know there're some accelerators for these processings, but I still need this pipeline to run some customzaiton algorithms, the basic processing is commonly like: filter, lookup-table, 2d-arrays operations(+,-,*)

    Thanks,

    Chris

  • Hi Chris,

    I am reading two key questions in your note. One is on the usage of the GLSL/GLES itself and the second is on setting up a OpenVX pipeline/graph.

    GLES should be available on QNX OS. On the OpenVX graph setup, let me check on that and get back.

    Thanks,

    Gowtham

  • Hi Gowtham,

    The 2 key questions you mentioned is extractly I'm expected, Thank you and waiting your reply, it couldn't be better if you can provide a example and comment for it.

    Thanks,

    Chris

  • Hello Chris,

    TDA4VM SoC has a HW ISP which has support for full image processing pipeline, including local tone mapping. It will be more efficient to use the ISP instead of GPU for image processing.

    Regards,

    Mayank

  • Hi Chris,

    See below reference to some documentation on the usage of OpenVX for developing applications using HW accelerators.

    Thanks,

    Gowtham