DefectDetect: A lightweight application for manual image annotation and patch extraction

This paper presents DefectDetect, a lightweight desktop tool for manual image annotation and automatic patch extraction, developed to assist in creating annotated datasets for machine learning problems. Although it is initially designed for annotating leather defects, the application supports broader use cases. Users can annotate freely, automatically extract smaller image patches and corresponding binary masks with adjustable stride, selected defect types and rating (0–2), and export data in PNG, JSON, YOLO, or Pascal VOC formats. The tool is fully GUI-based, requires no coding knowledge for usage, and supports session saving and batch image loading. The application has proven effective in academic contexts through its use in research activities. Future plans include the addition of shape annotation functionality and support for batch processing. • Lightweight GUI tool for manual image annotation and automated patch extraction. • Flexible export to PNG, JSON, YOLO, and Pascal VOC for easy ML/DL integration. • User-defined ratings, customizable patch sizes, and stride for dataset generation. • Designed for efficient preparation of annotated datasets from high-resolution images. • Potential applicability in diverse domains requiring image-based dataset creation.