yscv-cli
Command-line tool for real-time inference (detection + tracking + recognition pipeline), camera diagnostics, and dataset evaluation. Argument-driven (no subcommands).
# Enumerate camera devices (filtered by optional name match)
# Run camera diagnostics with a frame budget and JSON report output
# Evaluate detection predictions against COCO ground truth
# Default invocation runs the live pipeline (detect → track → recognize)
Modes
| Flag | What it does |
|---|---|
--list-cameras |
Enumerate camera devices, optionally filtered by name |
--diagnose-camera |
Run camera capture diagnostics, write JSON report |
--validate-diagnostics-report <path> |
Validate a previously saved diagnostics report against thresholds |
--eval-detection-{jsonl,coco,openimages,yolo,voc,kitti,widerface}-* |
Run detection evaluation against the named dataset format |
--eval-tracking-{jsonl,mot}-* |
Run tracking evaluation (MOTA / MOTP / IDF1 / HOTA) |
| (default) | Live pipeline: camera capture → detection → tracking → optional recognition, with benchmarking and event logging |
Features
[]
= [] # Live camera input
Configuration
JSON-based config for thresholds, model paths, and pipeline options.
Tests
42 tests covering config parsing, pipeline setup, evaluation output.