Video annotation tool

The tool enables researchers to label objects in video frames using bounding boxes, polygons, or polylines, with support for frame-level tracking and short-term propagation. It operates offline on local data, enabling privacy and reproducibility. Its open-source nature allows custom exporters and plugin extensions, reducing vendor lock-in.
Conceptually, users ingest a video, define a labeling schema, and annotate frames through an editable canvas that records spatial annotations and temporal associations. Annotations can be exported to standard datasets such as COCO, YOLO, or Pascal VOC, enabling seamless integration with ML pipelines. It emphasizes data integrity through unique IDs, coordinate validation, and format-conformant outputs.
Who benefits: AI researchers, data-labeling teams, robotics engineers, and video analytics practitioners who require transparent tooling, reproducible results, and control over data handling. Unique value lies in its open-source core, extensible exporters, offline operation, and support for multi-type annotations with tracking across frames. Core features include robust schema management, multiple annotation types, quality checks, and standard-format exports. Typical use cases cover surveillance labeling, autonomous systems dataset creation, and media analytics experiments.
Use cases span initial labeling, iterative model training, and dataset auditing, where traceability and reproducibility are essential.