• State Resolved
  • Locked Locked
  • Replies 1 reply
  • Subscribers 98 subscribers
  • Views 1045 views
  • Users 0 members are here
Support feedback
Options
Options
Related

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: openvx convert vx_tensor to vx_image

Ayman Khalaf1
Prodigy 230 points
Part Number: TDA4VM

Hi 

In the following demo http://software-dl.ti.com/jacinto7/esd/processor-sdk-rtos-jacinto7/latest/exports/docs/vision_apps/docs/user_guide/group_apps_dl_demos_app_tidl_seg.html
I want to just display the output of the segmentation model without combining it with the input image.

to do that I need to convert the output of the model which is of type vx_tensor to vx_image. I checked the documentation of openvx and found a function called vxCreateImageObjectArrayFromTensor.
Documentation link: https://www.khronos.org/registry/OpenVX/specs/1.2/html/d7/d0f/group__group__object__tensor.html#ga64427e7cf4935ceb204a18b8675dbbe3

however, I couldn't find it in the tiovx library. So how can I convert from vx_tensor to vx_image?

Regards.

  • 0
    TI__Genius 10355 points

    Hi,

    On TDA4x we support OpenVx 1.1 spec and not OpenVx 1.2 so you will not find vxCreateImageObjectArrayFromTensor api.

    Also the output of TIDL in app_tidl_seg will be an 8 bit plane where each value will be class-id's. So if you have 10 classes it will be from 0 to 9. 

    You will not be able to "display" this on the screen, unless you scale it appropriately. 

    If your intention is to just access the TIDL output then you have 2 options,

    1. Make the TIDL node output tensor as a graph parameter and enqueue/dequeue similar to scaler node's input imgae.

    2. You can attach a companion fileio node to TIDL node which accepts the output tensor as an argument and execute the node on A72. One example of this is done in app_multi_cam where we have tapped intermediate outputs of each node in the pipeline. You can take a look at vision_apps/kernel/fileio module and look at writing tensor data formats to media.

    Regards,
    Shyam