yscv-cli 0.1.8

CLI for inference, benchmarking, and evaluation
Documentation

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)
cargo run -p yscv-cli --features native-camera -- --list-cameras

# Run camera diagnostics with a frame budget and JSON report output
cargo run -p yscv-cli --features native-camera -- \
    --diagnose-camera \
    --diagnose-frames 120 \
    --diagnose-report report.json

# Evaluate detection predictions against COCO ground truth
cargo run -p yscv-cli -- \
    --eval-detection-coco-gt gt.json \
    --eval-detection-coco-pred pred.json \
    --eval-iou 0.5 \
    --eval-score 0.0

# Default invocation runs the live pipeline (detect → track → recognize)
cargo run -p yscv-cli --features native-camera -- \
    --camera --device 0 --detect-target people --detect-score 0.4

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

[features]
native-camera = []  # Live camera input

Configuration

JSON-based config for thresholds, model paths, and pipeline options.

Tests

42 tests covering config parsing, pipeline setup, evaluation output.