---

name: industrial-pipelines description: "Configures PaddleDetection PP-Human, PP-Vehicle, PP-Tracking, ReID, plate, attribute, action, violation, and video/RTSP industrial pipelines." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Industrial Pipelines

Use this route when the task names PP-Human, PP-Vehicle, PP-Tracking, pedestrian or vehicle analysis, ReID/MTMCT, tracking, attribute recognition, plate recognition, action/fall/fight/smoking/phoning/intrusion detection, illegal parking, vehicle press-line/retrograde, image/video/RTSP input, or deploy/pipeline.

Workflow

1. Choose a pipeline family: PP-Human, PP-Vehicle, or PP-Tracking-like MOT deployment.
2. Summarize the selected YAML with scripts/summarize_pipeline_config.py. Confirm enabled modules, model directories, visual output, and input type.
3. Stage local model directories before running. Many pipeline configs use URLs/auto-downloads; treat them as explicit network operations.
4. Choose one input mode: --image_file, --image_dir, --video_file, --video_dir, --rtsp, or --camera_id. Avoid specifying multiple modes unless testing parser priority.
5. Start CPU/paddle mode on a short local image/video fixture. Promote to GPU/TensorRT or multi-camera only after model/input preflight passes.
6. For counting or region-based tasks, set --region_type, --region_polygon, and related flags deliberately and record the coordinate system.

References

• references/pipeline-configuration.md: parser flags, config override semantics, module names, and input priority.
• references/pphuman-ppvehicle-workflows.md: PP-Human and PP-Vehicle feature routing.
• references/troubleshooting.md: pipeline dependency, model download, video, RTSP, tracking, and region errors.