Examples

End-to-end recipes covering the most common deployment patterns.

Single Stream with Display

Detect objects in a video file and show results on screen.

pipeline:
  name: "Display Demo"

streams:
  - uri: "file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h264.mp4"

models:
  - name: "yolo-detector"
    type: "detector"
    architecture: "yolo"
    onnx_file: /path/to/model.onnx
    label_file: /path/to/labels.txt

sinks:
  width: 1280
  height: 720
  outputs:
    - type: "display"

Single Stream with Kafka Analytics

Run detection and send results to Kafka (no video output).

pipeline:
  name: "Kafka Analytics"

streams:
  - uri: "file:///path/to/video.mp4"
    source_id: 0

models:
  - name: "yolo-detector"
    type: "detector"
    architecture: "yolo"
    onnx_file: /path/to/model.onnx
    label_file: /path/to/labels.txt

  - type: "tracker"
    tracker_type: "NvSORT"

sinks:
  width: 1920
  height: 1080
  outputs:
    - type: "fps"
    - type: "kafka"
      broker: "kafka-host:9092"
      topic: "detections"

Multi-Stream with Tiled RTSP Output

Two cameras combined into a single tiled RTSP stream.

pipeline:
  name: "Tiled Output"

streams:
  - uri: "rtsp://camera1/live"
    source_id: 0
  - uri: "rtsp://camera2/live"
    source_id: 1

models:
  - name: "yolo-detector"
    type: "detector"
    architecture: "yolo"
    onnx_file: /path/to/model.onnx
    label_file: /path/to/labels.txt

  - type: "tracker"
    tracker_type: "IOU"

sinks:
  width: 1280
  height: 720
  outputs:
    - type: "rtsp"
      url: "rtsp://localhost:8554/tiled"
      codec: "h264"
      bitrate: 4000000

Multi-Stream with Per-Stream Outputs

Two cameras, each with its own RTSP output plus a tiled display.

pipeline:
  name: "Per-Stream Output"

streams:
  - uri: "rtsp://camera1/live"
    source_id: 0
  - uri: "rtsp://camera2/live"
    source_id: 1

models:
  - name: "yolo-detector"
    type: "detector"
    architecture: "yolo"
    onnx_file: /path/to/model.onnx
    label_file: /path/to/labels.txt

  - type: "tracker"
    tracker_type: "NvSORT"

sinks:
  width: 1280
  height: 720
  outputs:
    - type: "display"
    - type: "rtsp"
      mode: "multi"
      codec: "h264"
      bitrate: 4000000
      streams:
        - source_id: 0
          url: "rtsp://localhost:8554/cam0"
        - source_id: 1
          url: "rtsp://localhost:8554/cam1"

Detector + Classifier Chaining

Detect vehicles, track them, then classify their color.

pipeline:
  name: "Chaining Pipeline"

streams:
  - uri: "file:///path/to/video.mp4"
    source_id: 0

models:
  # Primary detector
  - name: "vehicle-detector"
    type: "detector"
    architecture: "yolo"
    onnx_file: /path/to/vehicle_model.onnx
    label_file: /path/to/vehicle_labels.txt
    active_classes: [0, 2]

  # Tracker between detector and classifier
  - type: "tracker"
    tracker_type: "IOU"

  # Secondary classifier on detected objects
  - name: "color-classifier"
    type: "classifier"
    onnx_file: /path/to/color_model.onnx
    label_file: /path/to/color_labels.txt
    operate_on: "vehicle-detector"
    operate_on_classes: [0, 2]

sinks:
  width: 1280
  height: 720
  outputs:
    - type: "display"