Skip to main content

DeepStream

This guide provides instructions for installing and running the NVIDIA DeepStream SDK on Jetson Orin devices.DeepStream SDK is a comprehensive streaming analytics toolkit built on GStreamer, designed for real-time AI-based multi-sensor processing including video, audio, and image analytics. It enables GPU-accelerated video analysis pipelines and is highly optimized for the CUDA/NvMedia architecture on Jetson platforms.

1. Overview​

  • Real-time video analytics SDK provided by NVIDIA
  • Optimized with TensorRT and CUDA for maximum performance
  • Supports multi-stream AI inference and object tracking
  • Input sources include RTSP, USB/CSI cameras, and local video files
  • Built-in support for object detection, classification, and tracking

This guide covers：

  • Installation methods (via .deb packages and Docker)
  • Running sample DeepStream pipelines
  • Integrating custom models (e.g. YOLO, SSD, etc.)
  • Docker-based deployment using jetson-containers
  • Troubleshooting and optimization tips

overview

2. System Requirements​

Hardware​

ModelMinimum Requirement
DeviceJetson Orin Nano / NX
Memory≥ 8GB
Storage≥ 10GB

Software​

  • JetPack 6.1 GA or later (L4T ≥ R36.4)
  • Ubuntu 20.04 / 22.04
  • CUDA、TensorRT and cuDNN（included with JetPack 中）
  • Docker(optional, for containerized deployment)

3. Installation DeepStream​

  • glib Migration： To migrate to a newer version of glib (e.g., 2.76.6), please follow these steps:

    sudo pip3 install meson
    sudo pip3 install ninja

    Build and Install glib：

    git clone https://github.com/GNOME/glib.git
    cd glib
    git checkout <glib-version-branch>
    # e.g. 2.76.6
    meson build --prefix=/usr
    ninja -C build/
    cd build/
    sudo ninja install

    Confirm GLib Version：

    pkg-config --modversion glib-2.0

  • Install Required Libraries：

    sudo apt update
    sudo apt install -y \
      libssl1.1 \
      libgstreamer1.0-0 \
      gstreamer1.0-tools \
      gstreamer1.0-plugins-good \
      gstreamer1.0-plugins-bad \
      gstreamer1.0-plugins-ugly \
      gstreamer1.0-libav \
      libgstrtspserver-1.0-0 \
      libjansson4 \
      libyaml-cpp-dev

  • Install librdkafka (for Kafka Protocol Adapter)

  1. Clone the librdkafka repository from GitHub：
git clone https://github.com/confluentinc/librdkafka.git
  1. Configure and build the library:
cd librdkafka
git checkout tags/v2.2.0
./configure --enable-ssl
make
sudo make install
  1. Copy the compiled libraries to the DeepStream directory:
sudo mkdir -p /opt/nvidia/deepstream/deepstream/lib
sudo cp /usr/local/lib/librdkafka* /opt/nvidia/deepstream/deepstream/lib
sudo ldconfig

Method 1: Installation via SDK Manager​

  1. Download and install SDK Manager from NVIDIA’s official website .

  2. Connect the Jetson Orin device via USB-C to PC.

  3. Launch SDK Manager：running sdkmanager in the host and log in with your NVIDIA Developer account.

  4. Select hardware and JetPack version in SDK Manager.

  5. Check DeepStream SDK in "Additional SDKs".

  6. Follow on-screen instructions to complete installation.

Method 2: Using DeepStream Tar Package​

  1. Download the DeepStream SDK tar from the NVIDIA DeepStream Download Page（Example deepstream_sdk_v7.1.0_jetson.tbz2)

  2. Extract and install：

sudo tar -xvf deepstream_sdk_v7.1.0_jetson.tbz2 -C /
cd /opt/nvidia/deepstream/deepstream-7.1
sudo ./install.sh
sudo ldconfig

Method 3: Using DeepStream Debian Package​

  1. Download the Debian Package from DeepStream Debian Download page（Example:deepstream-7.1_7.1.0-1_arm64.deb)

  2. Install the package:

sudo apt-get install ./deepstream-7.1_7.1.0-1_arm64.deb

Method 4: Using DeepStream Docker​

  1. Install Docker and NVIDIA Container Toolkit.

  2. Pull DeepStream container：

docker pull nvcr.io/nvidia/deepstream-l4t:6.1-samples
  1. Run DeepStream container：
docker run -it --rm --runtime=nvidia \
  -v /tmp/.X11-unix:/tmp/.X11-unix \
  -e DISPLAY=$DISPLAY \
  nvcr.io/nvidia/deepstream-l4t:6.1-samples

(Optional) Use the jetson-containers community projec jetson-containers：

jetson-containers run dusty-nv/deepstream

Verification​

Check Version：

deepstream-app --version-all

Expected Output：

 deepstream-app version 7.1.0
 DeepStreamSDK 7.1.0
 CUDA Driver Version: 12.6
 CUDA Runtime Version: 12.6
 TensorRT Version: 10.3
 cuDNN Version: 9.0
 libNVWarp360 Version: 2.0.1d3

4. Running Examples​

Step 1: Run the Default Reference App​

  1. Navigate to the built-in sample configuration directory:
cd /opt/nvidia/deepstream/deepstream-7.1/samples/configs/deepstream-app
  1. Run the reference application:
# deepstream-app -c <path_to_config_file>
deepstream-app -c source30_1080p_dec_infer-resnet_tiled_display_int8.txt

This command launches a tiled display showing real-time object detection results from multiple video streams:

deepstream_app_5x8​

Step 2: Use USB or CSI Camera​

Modify the[source0]section of the configuration file to enable camera input:

[source0]
enable=1
type=1
camera-width=1280
camera-height=720
camera-fps-n=30

Run the app with your updated config:：

deepstream-app -c <your_camera_config>.txt

🎥 For DeepStream configuration：USB camera usestype=1，CSI camera uses GStreamer source element nvarguscamerasrc

Step 3：Use RTSP Stream as Input​

To connect to an IP camera stream, update the source block：

[source0]
enable=1
type=4
uri=rtsp://<your-camera-stream>

Step 4: Run Sample Application​

Navigate to the sample app directory：

cd /opt/nvidia/deepstream/deepstream-7.1/sources/apps/sample_apps/deepstream-test1

Compile the source code:

sudo make CUDA_VER=12.6

Run the application:

./deepstream-test1-app dstest1_config.yml

deepstream_od

For more sample source code, refer to: /opt/nvidia/deepstream/deepstream/sources

5. Integrating a Custom Model​

DeepStream supports custom model integration using TensorRT or ONNX formats.

Step 1: Convert the Model to TensorRT Engine​

Usetrtexec or tao-converter to convert your ONNX model:

trtexec --onnx=model.onnx --saveEngine=model.engine

Step 2: Update DeepStream Configuration File​

Edit the model configuration section:

[primary-gie]
enable=1
model-engine-file=model.engine
network-type=0

For more DeepStream + TAO Toolkit integration examples, refer to: https://github.com/NVIDIA-AI-IOT/deepstream_tao_apps

6. Additional Examples​

deepstream_python_apps deepstream_python

7. Troubleshooting​

IssueSolution
No image display in DockerMount the X11 socket and set the DISPLAY environment variable
Low frame rateUse INT8 engine format or reduce input video resolution
USB camera not detectedRun v4l2-ctl --list-devices to verify the device
GStreamer errorsEnsure all required plugins are installed; reflash JetPack if necessary
RTSP stream lag or frame dropsSet drop-frame-interval=0 or latency=200

8. Appendix​

Key Paths​

PurposePath
Sample config files/opt/nvidia/deepstream/deepstream/samples/configs/
Model engine files/opt/nvidia/deepstream/deepstream/models/
Log directory/opt/nvidia/deepstream/logs/
DeepStream CLI tool/usr/bin/deepstream-app

References​