Place object detector on jetson tx2

I’m going to place a real-time object detector on the jetson tx2.

I followed the object detection tutorial, but I don’t know the process after the training.

So I asked this question, and I got the answer that I prefer using deepstream sdk rather than that method.

But I first heard a deepstream sdk.

1.We trained with object detection tutorial on sensorflow to obtain .ckpt files. Can you tell me the following process for placing object detectors in jetson?
2. Do you have a tutorial on deepstream sdk?
3. Can I train with my data with deepstream sdk and apply it to object detection?

Hi,

  1. Workflow should like this: .pb → uff → TensorRT engine.
    For more detail, please check another topic for information:
    Real-time object detection with jetson tx2 - #3 by AastaLLL

  2. Here are some deepstream document for your reference:
    NVIDIA Metropolis Documentation
    A deepstream sample for TensorFlow detection model can be found here:
    /opt/nvidia/deepstream/deepstream-4.0/sources/objectDetector_SSD/

  3. Deepstream is more focusing on inferencing.
    For training a model, please use our transfer learning toolkit.
    https://developer.nvidia.com/transfer-learning-toolkit

Thanks.