humansignal/labelimg

LabelImg is a graphical tool for annotating images to train object detection models. Before a machine learning system can learn to recognize objects in photos, say, cats, cars, or people, it needs thousands of example images where humans have manually drawn boxes around those objects and labeled what they are. LabelImg provides a desktop application that makes this process straightforward: you open a folder of images, draw rectangular boxes around objects, name them, and save the results. It saves annotations in three standard formats: PASCAL VOC (an XML-based format used by the ImageNet dataset), YOLO (a compact text format used by the YOLO family of object detection models), and CreateML (Apple's format). This makes LabelImg compatible with the most popular training pipelines without needing to convert files. You'd use this if you're building a custom image recognition or object detection model and need to create labeled training data from scratch, or if you're a researcher annotating a dataset for academic work. It runs on Linux, macOS, and Windows and installs simply via pip. Note: LabelImg is no longer actively developed, the repository now points users to Label Studio, a broader open source annotation platform by HumanSignal that handles images, text, audio, video, and time-series data. LabelImg still works for its original purpose but receives no new updates. Written in Python using Qt for its graphical interface.