DeepStream deployment
Support NVIDIA DeepStream SDK 6.1.1 / 6.1 / 6.0.1 / 6.0 docker images.
Getting started
Suported models
Benchmarks
Config
device = NVIDIA RTX3060 12GB, A100
batch-size = 1
eval = val dataset, eval dataset
sample = 1920x1080 video and image
Results
NOTE: IoU=0.5, FPS = RTX3060 ; A100
| DeepStream | Resolution | Val set | Eval set | FPS (with display) |
|---|---|---|---|---|
| yolov4 | 608 | 0.865 | 0.824 | |
| yolov4-fp32 | 608 | 40 ; 75 | ||
| yolov4-fp16 | 608 | 85 | ||
| yolov7 | 640 | 0.958 | 0.960 | |
| yolov7-fp32 | 640 | 50 ; 86 | ||
| yolov7-fp16 | 640 | 140 |
Docker usage
-
x86 platform
nvcr.io/nvidia/deepstream:6.1.1-devel nvcr.io/nvidia/deepstream:6.1.1-triton -
Jetson platform
nvcr.io/nvidia/deepstream-l4t:6.1.1-samples nvcr.io/nvidia/deepstream-l4t:6.1.1-triton
If watch realtime video, run
xhost +
before run container
docker run --gpus all -it --rm --net=host --privileged -v path-to-this-repo:/opt/nvidia/deepstream/deepstream-6.1/sources/DeepStream-Yolo -e DISPLAY=$DISPLAY -w /opt/nvidia/deepstream/deepstream-6.1 nvcr.io/nvidia/deepstream:6.1.1-devel
Inside container, install requirements
/opt/nvidia/deepstream/deepstream/user_additional_install.sh
wget https://github.com/NVIDIA-AI-IOT/deepstream_python_apps/releases/download/v1.1.4/pyds-1.1.4-py3-none-linux_x86_64.whl
pip3 install pyds-1.1.4-py3-none-linux_x86_64.whl
cd sources/team-winner-is-coming
pip3 install -e requirements.txt
Check cuda version by nvcc --version and build customparser C++ function
export CUDA_VER=11.7
cd nvdsinfer_custombboxparser
make
cd ..
cd nvdsinfer_custom_impl_Yolo
make
cd ..
Then run deepstream app
python3 deepstream_app.py