Composer- loading different object detection models

Accelerated Computing Intelligent Video Analytics DeepStream SDK
,
1

Please provide complete information as applicable to your setup.

• Hardware Platform: dGPU RTX3060
**• DeepStream Version: 6.1.0
• JetPack Version (valid for Jetson only)
• TensorRT Version: 8.2.5.1
• NVIDIA GPU Driver Version: 510.47.03
• Composer 2.0.0

Everything configured as per Quickstart Guide — DeepStream 6.1.1 Release documentation

Hello,

I am just getting started with Composer- so apologies if this is an obvious answer

How do I use object detection models other than the standard ones available with the samples ??

Specifically, I would like to use PeopleNet- but it does not appear in the list
This screenshot is from my PC

screenshot composer
screenshot composer1140×768 64.2 KB

Whereas in the tutorial Creating an AI Application — DeepStream 6.1.1 Release documentation
other models (including PeopleNet) are shown

screenshot tuturial
screenshot tuturial1322×868 115 KB

I’ve tried modifying config files such as /opt/nvidia/deepstream/deepstream/samples/configs/deepstream-app/config_infer_primary.txt
and I have also run /opt/nvidia/deepstream/deepstream/samples/prepare_ds_triton_model_repo.sh

and followed the steps for the tao pretrained models

Screenshot from 2022-07-17 13-48-36
Screenshot from 2022-07-17 13-48-36935×902 41.1 KB

thank you

Andrew

3

Can you refer to the trouble shooting page first? Troubleshooting — DeepStream 6.1.1 Release documentation

4

thank you for the reply- but I have already gone through this, and other documentation

Is there a specific item in the troubleshooting page?- I can’t see anything that relates to this issue

Andrew

5

Just bumping this

I’ve spent a fair bit of time trying various config files- with no success

Andrew

7

Final attempt

I would be very grateful if someone from NVIDIA (or anyone else) can assist

thank you

Andrew

8

The models you mentioned are TAO models. Currently TAO models are not integrated in Graph Composer by default.

9

Hi Fiona

thanks for the reply

Does this mean that the models* shown in the NVIDIA documentation (second screenshot above) can’t be used ?

*such as PeopleNet

If they are not integrated ‘by default’- how can they be integrated or loaded ??

Andrew

12

We will consider to put a complete sample in the package in the future.

14

I am also trying to write/create my own decoder in integration with ROS for a different AI model which is not from Nvidia NGC catalog and I would also like some documentation/support on how we can deploy our own custom models. That will be really helpful.