Installing Deepstreak SDK 6.0 on JetPack 4.6 - Help

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Using Jetson Nano 2GB
Need to install Deepstream SDK 6.0 on my Jetson w/JetPack 4.6
I have seen the installation guides and tried to follow along the best I can but my installation failed and dont know how to uninstall it and go back, so I have a fresh install of JetPack 4.6 on my SD card now and would like to know the steps involved to install Deepstream 6.0 on my Jetson. Please please help. I have been trying to figure this out for 2 + weeks now and can’t get any further, I’ve tried following these steps: Quickstart Guide — DeepStream 6.1.1 Release documentation

It was giving me errors when I would try to run a sample application from the folder, and I am completely lost at this point and really need to figure this out to help project I’m working on to show my colleagues. If anyone could give me clearer instructions on what to do or what I need to do I would be very grateful. I’ve had no trouble following dusty-nv’s instructions for simple object detection but need something more robust capable of counting people etc. Thanks in advance.

3

Sorry for the late response, have you managed to get issue resolved or still need the support? Thanks

4

Yeah, I found the course you can take through NVIDIA to install it using it along with Jupyter Labs. Hopefully I can get something working soon, still have no idea how to train my own model though with deepstream (taking pictures and adding bounding boxes). Do you know if its possible to use the data capture tool in JetPack to do this or do I need to do something different? Sucks there aren’t many tutorials out there, only showing what its capable of (which is amazing btw). Thank you for the reply though, if you can help or give me any guidance on how to train my own model using SSD-mobilenet-v2 or something like that, would be great. Thank you!@@

6

Hi,

Deepstream is a library for inference.
If you want to train a model, please check our TAO toolkit:
https://developer.nvidia.com/tao

More, please apply training on a desktop GPU.
Nano is designed for inference and has extremely limited resources which are not suitable for training.

Thanks.

7

Hello, thanks for your reply. Can I train it on my desktop GPU then transfer it to my jetson to run the object detection? Is that possible? Thank you again for your advice, I really appreciate it.

8

Hi,

YES. But please noted that you need to create the TensorRT engine file from Jetson directly.
Below is our recommended development cycle for your reference:

Thanks.

10

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.