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.

WEBENCH® Tools/TDA3MD: About TI TDA Vision SDK framework design , eg. usecases : vip_single_cam_object_detection2

Part Number: TDA3MD

Tool/software: WEBENCH® Design Tools

Hello Experts,

Recently, I'm doing some research base on TDA3 (PROCESSOR_SDK_VISION_03_07_00_00),and i have already read your docs in vision_Sdk.

But there is still some understand about framework design or usecase design as below. If anybody can help? Thanks a lot!

  1. As the usecase (vip_single_cam_analytics2) Demo shows,Alg_ImgPyramid duplicates information to DSP1 Links and EVE1 Links,But DSP1 Links didn't process anything and be merge to ObjectDetect links. So ,What is the design intents?or just keep one link is same effect?
  2. In my Application,both EVE and DSP is needed, EVE and DSP need to keep communication with a little data to swap.  Is it the best way that use Shared DDR Memory pointer in vision  Architecture?
  3. About DUP/Sync/Merge··· Link,Can you show me more rules of use or need to be keep noticed.

  • Hello Experts,

    Recently, I'm doing some research base on TDA3 (PROCESSOR_SDK_VISION_03_07_00_00),and i have already read your docs in vision_Sdk.

    But there is still some understand about framework design or usecase design as below. If anybody can help? Thanks a lot!

    1. As the usecase (vip_single_cam_object_detection2) Demo shows,Alg_ImgPyramid duplicates information to DSP1 Links and EVE1 Links,But DSP1 Links didn't process anything and be merge to ObjectDetect links. So ,What is the design intents?or just keep one link is same effect?
    2. In my Application,both EVE and DSP is needed, EVE and DSP need to keep communication with a little data to swap.  Is it the best way that use Shared DDR Memory pointer in vision  Architecture?
    3. About DUP/Sync/Merge··· Link,Can you show me more rules of use or need to be keep noticed.

  • Hi,

    Sorry i did not understand your 1st question.

    For 2nd question yes you can use Shred ddr region for communication between 2 cores.

    Please refer ipc in/out link for more information

    Q3) Dup, merge & sync these are generic link.

    Dup is used to duplicate the the output of a link and provide the same to multiple links

    Sync is used to synchronize the frames of all channels.

    Merge is used to receive input from multiple input links and coming all into single output que and provide it to next link.

    Regards,

    Anuj

  • Hi Anuj,

    sorry,there is some clarification is needed.

    My first question in detail as below:

    At the TDA3 SDK(PROCESSOR_SDK_VISION_03_07_00_00),I find a usrcase about object detection, in (~\vision_sdk\apps\src\rtos\usecases\vip_single_cam_object_detection2). But I have some understand at the chains design.

    please refer to the attachment, this is a object detection DMEO form TI designer. 

    1. As the picture shows, Alg_ImgPyramid duplicates information to DSP1 Links and EVE1 Links,But DSP1 Links didn't process anything and be merge into ObjectDetect links. So ,What is the design intents?or just keep one link is same effect?

    2. As the picture shows, The Links Dup_capture split in two Links.And the Q0 straightforward to Merge_alg, I Can't understand the design intents also.

    3. So,I'm a little confused that did i need to do like this in my application.

    Thanks and regards,

    Nick

  • Hi,

    You can ignore ipc links in the image.

    If the data is transferred from 1 core to other then ipc link will come into picture.

    As the data from Dup_imagePyramid running on ipu1_0 is going to Merge_objectdetect running on DSP1. So there will be ipc links.

    And we are using dup link to make 2 copyies of orignal data  from Alg_imagePyramid out of which 1 goes directly to Alg_ObjectClassification without any processing and other also goes to same link but after some processing.

    There are some link which required 2 input, 1 original video data and 1 some processed data so in that case we use dup to split and merge to merge data.

    Regards,

    Anuj