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.

PROCESSOR-SDK-TDAX: Framerate Improvement in Vision SDK 3.02 with TDA3x

Part Number: PROCESSOR-SDK-TDAX

Hi Champs

There's a request that Customer wants to get achieved their goal.

They're using the TDA3x Vision SDK 3.02 version.

The usercase is the object detection in vip_single_cam_object_detection2.

The max resolution is 1280x720, but the frame rate is about 14 to 17 frames.

We would like to ask if we can improve the performance so that we can get about 30 frames on 1280x720.

Please help this issue to improve the framerate with object detection.

Regards, 

Jack

  • Hi Jack,

    Below couple of options they can try to improve the frame rate,

    1. Increase the confidence threshold for detection outptut layer by updating below 2 parameters, and then reimport the model to get new bin files

    •    keep_top_k: 20
    •     confidence_threshold: 0.15

    2. Make sure detection output layer runs on the DSP and rest all layers on the EVE to get optimal performance by  setting proper "layersGroupId” while importing the model, please refer to FAQ 21 and 22 in the TIDL user guide to get more details about this parameter.

    Thanks,

    Praveen