💡 Examples

The examples below demonstrate pyppbox configuration, detection, tracking, ReID, evaluation, and optional modules. Install the package and required model/data wheels first; see 🚀 Getting Started.

Examples 1–14 use the standalone/direct-module interfaces; example 12 also includes the V4 ArcFace/AdaFace reiders. Examples 15–18 cover additional V4 workflows. Install each example’s optional requirements from the repository root before changing directories.

All example scripts use the repository’s examples/ directory as their working directory. From the repository root:

cd examples
python example_01_basic.py

For the newer command-line examples, use the same working directory:

python example_15_rfdetr.py --device cpu --frames 30
python example_16_export_adaface_onnx.py --help
python example_17_benchmark_faces.py --help
python example_18_onnx_face_reid.py --help

Relative data/, cfg/, single_config/, configuration-file, and output paths resolve from this working directory. Examples 3, 4, 5, 11, and 13 contain C:/pyppbox_v3/ placeholders in their scripts or custom configs; replace these with your installed pyppbox directory. Find it with:

python -c "import os, pyppbox; print(os.path.dirname(pyppbox.__file__))"

Example 12 retrains the FaceNet, Torchreid OSNet-AIN, Torchreid OSNet, ArcFace, and AdaFace GTA V classifiers. Example 18 trains and recognizes a face using a custom ArcFace/AdaFace configuration. Model requirements are documented in Supported ReIDers. Device settings and automatic selection are described in Devices.

ℹ️ Source code and required input files are available here: {pyppbox repo}/examples.

• Example 1: Use some basic standalone functions
• Example 2: Use pyppbox’s internal config directory
• Example 3: Set a custom config directory
• Example 4: Set the main detector/tracker/reider by a YAML or JSON file
• Example 5: Set the main modules by a raw/ready dictionary
• Example 6: Set the main modules by names
• Example 7: Use result tools
• Example 8: Use real-time evaluation tools
• Example 9: Compare a generated result to a reference or ground-truth
• Example 10: Use GT (Ground-truth) as a detector
• Example 11: Train a classifier using a supported reider
• Example 12: Retrain GTA V classifiers for all supported reiders
• Example 13: Use a supported module directly
• Example 14: Use pyppbox in multithreading application
• Example 15: Use RF-DETR detection and SORT tracking
• Example 16: Export an AdaFace embedding model to ONNX
• Example 17: Compare FaceNet, ArcFace, and AdaFace
• Example 18: Recognize faces and train an ArcFace or AdaFace classifier