Samples of Python and C GStreamer pipelines with NVIDIA DeepStream modules:
To run examples:
- Start RTSP server, for example: link.
- Prepare streaming, for instance, with ffmpeg:
$ bash stream_video.sh <video_address>- For Jetson platform clone the repository. (!Streaming on Jetson is quite computational intensive. It is better to stream via another computer.)
- For dGPU use Docker.
$ docker build -f ./dockerfiles/deepstream-5.0-20.07-devel.Dockerfile -t gst_deepstream5.0_cuda10.2:dgpu .
$ xhost +local:docker
$ docker run --name ds_test --gpus=all --runtime nvidia -e DISPLAY=$DISPLAY \
--rm -i -t --net=host \
-v $PWD:/home/ubuntu/inference-gstreamer \
-v /tmp/.X11-unix/:/tmp/.X11-unix "gst_deepstream5.0_cuda10.2:dgpu" bash
Examples:
| Example | Description | Support |
|---|---|---|
gst_read_rtsp.py |
RTSP stream reading with reconnection | Jetson, dGPU |
gst_read_multiple_rtsp.py |
Multiple RTSP streams reading with reconnection | Jetson, dGPU |
gst_primary_detector.py |
RTSP object detection with deepstream PrimaryModel | Jetson, dGPU |
gst_trafficcam_model.py |
TLT-pretrained TrafficCamNet model | Jetson, dGPU |
gst_dashcam_model.py |
TLT-pretrained DashCamNet model | Jetson, dGPU |
gst_multiple_rtsp_inference.py |
RTSP streams + DashCam + PeopleNet + Tracker | Jetson, dGPU |
Arguments parser:
-h, --help show this help message and exit
-ip ip str, rtsp ip address, default='127.0.0.1'
-port port int, rtsp port, default=8554
-name name str, rtsp address name or names, default='stream'
-codec codec str, video codec, default='h264
-debug_level debug_level str, GStreamer debug level, default=0
-d bool, display pipeline output
-v bool, more verbosity