This repository provides a collection of Python scripts and guidance for building a person detection system using YOLOv3. It targets researchers and engineers familiar with deep learning object detection who need to extract, format, and evaluate person-specific datasets for YOLOv3 training. The primary benefit is a streamlined workflow for dataset preparation and performance evaluation.

How It Works

The project leverages two main YOLOv3 implementations (AlexeyAB/darknet and pjreddie/darknet) and offers scripts to extract person-only images from VOC2007/2012 and COCO datasets. It includes utilities for converting VOC annotations to YOLO format and calculating Mean Average Precision (mAP). The scripts also facilitate training log visualization for monitoring progress.

Quick Start & Requirements

• Install: Clone AlexeyAB/darknet, download yolov3.weights.
• Prerequisites: OpenCV (version 3.40 recommended, 3.41 noted as problematic), CUDA (implied by darknet compilation).
• Dataset Preparation: Download VOC2007, VOC2012, and COCO datasets. Run extract_person.py and voc_label.py for dataset conversion.
• Training: Configure cfg/voc.data, data/voc.names, and cfg/yolov3-voc.cfg. Compile darknet with make. Train using ./darknet detector train ....
• Links: AlexeyAB/darknet, pjreddie/darknet

Highlighted Details

• Scripts for extracting person-class images from VOC2007/2012 and COCO datasets.
• Utilities for converting VOC annotations to YOLO format and calculating mAP.
• Guidance on visualizing training logs using tee and provided scripts.
• Supports both AlexeyAB/darknet and pjreddie/darknet YOLOv3 versions.

Maintenance & Community

No specific information on maintainers, community channels, or roadmap is provided in the README.

Licensing & Compatibility

The repository itself does not specify a license. It relies on the underlying darknet repositories, which have their own licenses (AlexeyAB/darknet is GPLv3, pjreddie/darknet is custom permissive). Compatibility for commercial use depends on the chosen darknet fork's license.

Limitations & Caveats

The project does not include YOLOv3 code or configuration files, requiring users to obtain them separately. Specific OpenCV version compatibility issues are noted. The README focuses on person detection, and adapting for other classes may require modifications.