Hello,

I only see on the official luxonis website that the following neural network models can be supported. (eg. TensorFlow, Caffe, ONNX, OpenVINO IR, and OpenVINO Model Zoo)

I would like to know if we can use Pytorch or PaddlePaddle to develop neural network models for deployment to OAK cameras?

I am looking forward to your reply. Thank you very much for your support!

  • erik replied to this.

    Hi Jalen_Xu ,
    Yep, Pytorch is actually the most popular way people train models that then run on OAK. Usual workflow is to convert pytorch to onnx, then to openvino, and then compile to blob. Example here. THoughts?

    Thanks, Erik

        erik Thanks for your reply.

        1. We would like to know if we can use PaddlePaddle to train models that then run on OAK. (PaddlePaddle is the open source deep learning platform launched by Baidu)

        2. Can we deploy recurrent neural network(RNN) models to OAK?

        3. Apart from Pytorch, TensorFlow, Caffe, ONNX, OpenVINO IR, and OpenVINO Model Zoo, what other AI frameworks does our OAK support to train models through?

        • erik replied to this.

            Hi Jalen_Xu ,

            1. Openvino has paddlepaddle conversion docs here.
            2. Should be possible. though might be limited. initial models for license plate recognition were using 2 LSTM cells
            3. Best to check what OpenVINO supports - in theory, all (popular) frameworks have uspport to convert to onnx.