Hi, Erik I have optimized FaceNet from OpenVino Open Model Zoo. It has some problems with input shape.

Is there any example for FaceNet deployment on OAK D!? Need Guidance.

    Hi SamiUddin
    As far as I'm aware there isn't an example for the FaceNet specifically. One quick tip which might solve the problem: Depends on what openvino version you are using, the blobconverter expects different input format

    2021.4 version here expects [Batch,Channels,Height,Width] input, while 2022.1 version expects [B,H,W,C] input

    Source

    Could you please paste the logs of the error you are getting?

    Thanks,
    Jaka

      Please have a look at. I have compiled faceNet model with blob converter. openvino 2022.1.

      Using face recognition demo of Depthai Experiments. Just changed model from arcface to facenet . and input from 112x112 to 160x160.

      Hi @SamiUddin

      1. Change the input size to 160x160 inside the script as well
      2. Data exceeds the range because it expects U8 but is fed FP16. Using legacy format doesn't help either.
        Instead, when compiling, use -U8 -il BCHW, first will set the model to U8 so it doesn't exceed range, and second will make sure the layer shape is correct.

      Hope this helps 🙂
      Jaka