Labelme

Inspired by MIT CSAIL's LabelMe and forked from pylabelme, this Python and Qt desktop app lets you outline objects with polygons, rectangles, circles, lines and points, tag whole images with flags for classification or cleanup, and step through video frames. Each image gets a sidecar JSON file, a format the maintainers keep stable, and bundled example scripts convert those files into VOC-style semantic or instance segmentation sets and COCO instance segmentation. Click a point and SAM or EfficientSAM proposes a polygon or mask, or type a prompt and YOLO-World or SAM3 suggests boxes. The interface is translated into about 20 languages and supports predefined labels, autosave and label validation via ~/.labelmerc. Version 7 moved to Qt6 through PySide6 and needs Python 3.12+ on 64-bit Windows, macOS or Linux, while the 6.3 line remains for Qt5 or older Python setups. Install with pip, a distro package, or a paid one-time standalone build from labelme.io that funds development. GPL-3.0 licensed, so redistributed modifications must stay open source. Best for researchers and small teams labeling images on their own machines.