Run TrafficCamNet on latest JetPack

[eugenyshcheglov]

• NVIDIA Jetson Xavier NX Developer Kit
• L4T 35.3.1

Hello.

I’m interested in how I can run the TrafficCamNet model on the latest versions of L4T. The instructions on the website describe running it using the deepstream-5.1-21.02-samples container, which is incompatible with my L4T version

Thank you.

[yingliu]

You can copy the config files from docker deepstream-5.1-21.02-samples (config_infer_primary_trafficcamnet.txt , deepstream_app_source1_trafficcamnet.txt , labels_trafficnet.txt) and use it in DeepStream 6.2, the etlt model and config files can be used in the latest DeepStream.

[eugenyshcheglov]

Okay, solved:

# docker pull nvcr.io/nvidia/deepstream-l4t:6.2-samples
# docker run --runtime=nvidia -it nvcr.io/nvidia/deepstream-l4t:6.2-samples /bin/bash
# ./user_additional_install.sh 
# apt install wget zip
# wget --content-disposition https://api.ngc.nvidia.com/v2/models/nvidia/tao/trafficcamnet/versions/pruned_v1.0.3/zip -O trafficcamnet_pruned_v1.0.3.zip
# cd /opt/nvidia/deepstream/deepstream/sources/apps/sample_apps/
# git clone https://github.com/NVIDIA-AI-IOT/deepstream_reference_apps.git
# cd /opt/nvidia/deepstream/deepstream/
# cp sources/apps/sample_apps/deepstream_reference_apps/deepstream_app_tao_configs/* samples/configs/tao_pretrained_models/
# cd /opt/nvidia/deepstream/deepstream/samples/configs/tao_pretrained_models/
# ./download_models.sh
# deepstream-app -c deepstream_app_source1_trafficcamnet.txt