DeepStream 8.0 Python Bindings — NVIDIA DGX Spark

Jupyter notebooks for building DeepStream 8.0 video analytics pipelines with Python on the NVIDIA DGX Spark.

Getting Started

Pre-built image

docker pull manasi1096/deepstream8-python:gb10
docker run --gpus all -it --rm -p 8888:8888 manasi1096/deepstream8-python:gb10 \
  bash -c "source /opt/ds-venv/bin/activate && jupyter notebook --ip=0.0.0.0 --port=8888 --no-browser --allow-root --NotebookApp.token='' --notebook-dir=/app/notebooks"

Open http://localhost:8888 in your browser.

OR

Build from source

git clone https://github.com/Manasi-NV/deepstream-spark-python-bindings.git
cd deepstream-spark-python-bindings
docker build -t deepstream-spark-python .
docker run --gpus all -it --rm -p 8888:8888 deepstream-spark-python

TensorRT engines are built automatically on first run (~5 min) and cached for subsequent runs.

Notebooks

Notebook	Description
Introduction_to_Deepstream_and_Gstreamer.ipynb	DeepStream SDK overview and GStreamer foundation concepts
Getting_started_with_Deepstream_Pipeline.ipynb	Plugin walkthrough and building a 4-class detection pipeline
object_detection.ipynb	Single-stream object detection with H.264 hardware encoding
object_detection_and_tracking_classification.ipynb	Detection + tracking + secondary classifiers (vehicle make & type)
object_detection_multistream.ipynb	Multi-stream (2x) detection with tiled side-by-side output

License

Creative Commons Attribution 4.0 International (CC BY 4.0)