How madhavan-hos/sdv-intel-edge-pipeline works
This is an SDV vision workshop stack: train or export a YOLO model on HPC, move an OpenVINO artifact to the edge, and combine RealSense obstacle-distance perception with FHD-camera lane detection in a local dashboard. A separate socket-based path supports remote HPC inference. The README documents model transfer and the major runtime choices; sampled code shows the sensor processing and remote request/response. Some setup and training details are represented only by documented capabilities, not inspected implementation.
Main entry points: local_sdv_pipeline_optimized.py Responsive dashboard · best_openvino_model OpenVINO model · edge_client_streamer.py Edge frame streamer · export_openvino.py OpenVINO export
Components 19
Model preparation
- YOLO training
session_05_model_hpc/train_yolo.py - OpenVINO export
session_05_model_hpc/export_openvino.py - Precision variants
session_05_model_hpc/optimization/export_openvino_precisions.py - OpenVINO model
models/best_openvino_model
Edge sensing
- Camera discovery
session_04_vision/camera_scanner.py - RGB-depth alignment
session_06_fusion/rgb_depth_fusion.py
Perception
- Lane detector
session_08_usecase/lane_detection.py - Obstacle detector
session_08_usecase/obstacle_detection.py - Device benchmark
session_07_acceleration/benchmark_devices.py
Integration paths
- Local dashboard
session_09_integration/local_sdv_pipeline.py - Responsive dashboard
session_09_integration/local_sdv_pipeline_optimized.py - Camera workers
session_09_integration/local_sdv_pipeline_optimized.py - Detection worker
session_09_integration/local_sdv_pipeline_optimized.py - Lane worker
session_09_integration/local_sdv_pipeline_optimized.py - HPC inference server
session_09_integration/hpc_server_inference.py - Edge frame streamer
session_09_integration/edge_client_streamer.py
Other components
- Workshop operator
- Edge cameras
- HPC host
Connections 26
- Workshop operator → YOLO training: trains model
README.md - YOLO training → OpenVINO export: exports weights
- OpenVINO export → OpenVINO model: writes model
- Workshop operator → OpenVINO export: exports weights
README.md - Workshop operator → Precision variants: exports variants
README.md - Workshop operator → OpenVINO model: transfers model
README.md - Camera discovery → Edge cameras: scans devices
session_04_vision/camera_scanner.py - Edge cameras → RGB-depth alignment: provides streams
- Edge cameras → Lane detector: provides frames
- Edge cameras → Obstacle detector: provides RGB-depth
session_08_usecase/obstacle_detection.py - RGB-depth alignment → Obstacle detector: provides aligned depth
session_06_fusion/rgb_depth_fusion.py - Workshop operator → Device benchmark: benchmarks devices
README.md - Workshop operator → Local dashboard: starts dashboard
README.md - Workshop operator → Responsive dashboard: starts dashboard
README.md - OpenVINO model → Local dashboard: loads model
README.md - OpenVINO model → Responsive dashboard: loads model
README.md - Responsive dashboard → Camera workers: starts workers
session_09_integration/local_sdv_pipeline_optimized.py - Responsive dashboard → Detection worker: starts workers
session_09_integration/local_sdv_pipeline_optimized.py - Responsive dashboard → Lane worker: starts workers
session_09_integration/local_sdv_pipeline_optimized.py - Camera workers → Detection worker: publishes frames
session_09_integration/local_sdv_pipeline_optimized.py - Camera workers → Lane worker: provides camera frames
session_09_integration/local_sdv_pipeline_optimized.py - Workshop operator → Edge frame streamer: starts remote mode
README.md - Edge cameras → Edge frame streamer: provides frames
session_09_integration/edge_client_streamer.py - Edge frame streamer → HPC inference server: sends frames
session_09_integration/edge_client_streamer.py - HPC inference server → Edge frame streamer: returns annotated frames
session_09_integration/hpc_server_inference.py - HPC host → HPC inference server: hosts inference