How to custom developing ml model in Jetson nano instead of jetson inference object detection problem?

[VK01]

How to custom developing ml model in Jetson nano instead of jetson inference object detection problem?

• I want to develop own deep learning model and deploy jetson nano board

Any suggestion?

[dusty_nv]

Hi @VK01, you can train your own object detection models in jetson-inference by following this part of the Hello AI World tutorial: https://github.com/dusty-nv/jetson-inference/blob/master/docs/pytorch-ssd.md

However, let’s say that you want to use your own custom network architecture. Some things you could do:

• use PyTorch to train the model, and just run the inference in PyTorch also on the Nano (you can install PyTorch on your Nano). You could use a library like torch2trt to accelerate it with TensorRT, without really changing your PyTorch code.

• use TensorFlow to train the model, and just run the inference in TensorFlow also on the Nano (you can install TensorFlow on your Nano). You could use a library like TF-TRT to accelerate it with TensorRT, without really changing your TensorFlow code.

• train your model using TAO Toolkit, and deploy it to your Jetson using DeepStream

Those are just a few of the ways that still give you the option of deploying it with TensorRT on your Jetson for higher performance.

[VK01]

Thanks! i will try this above mentioned ways