In this blog-post, we will download & run DeepStream YOLOR on NVIDIA Jetson TX2NX. First, we will download the YOLOR source files. Then, we will build & install the required packages. Finally, we will test the model file with DeepStream.

First of all, install deepstream using the command:

Now, download the YOLOR repository using the following command:

Upgrade pip setuptools wheel using the command:

Install numpy and matplotlib using the commands:

Edit the requirements.txt file in the yolor directory by changing
torch==1.7.0
torchvision==0.8.1
pycocotools==2.0
to 
torch>=1.7.0
torchvision>=0.8.1
pycocotools>=2.0

Then, install the requirements for Deepstream using the command:

Now, download the Deepstream-Yolo repository usinf the command:

Copy the gen_wts_yolor.py file from DeepStream-Yolo/utils directory to the yolor folder.

Download the yolor_p6.pt file from YOLOR repository using this link:

https://github.com/WongKinYiu/yolor

Now, move the yolor_p6.pt from the downloads directory to the yolor directory.

Generate the cfg and wts files using the command:

Copy the generated .cfg and .wts files in the yolor directory to the Deepstream-Yolo directory.

Now, cd (change directory) to the Deepstream-Yolo directory and compile the lib using the command:

Edit the config_infer_primary_yolor file in the DeepStream-Yolo directory as seen below and save.

Change 

network-mode=0

model-engine-file=model-b1_gpu0_fp32.engine

to

network-mode=2

model-engine-file=model-b1_gpu0_fp16.engine

Edit the deepstream_app_config.txt file as seen in the highlighted part and save.

Test the download model using the command:

You should obtain the results below after the command is successfully executed.

Thank you for reading our blog post.