videbat

Videbat is a cross-platform software framework designed to support the capture, organization, annotation, and analysis of video datasets for research and development in computer vision and related fields. The project emphasizes reproducibility, collaboration, and scalable processing workflows.

At its core, videbat consists of modules for data ingestion, annotation, preprocessing, and experiment management. The data ingest module can stream from cameras, IP cameras, video files, and existing repositories, while the annotation tools support bounding boxes, polygons, and keypoints.

An extensible processing pipeline enables augmentation, normalization, feature extraction, and model-integration steps, with support for GPU acceleration. The platform includes dataset versioning and experiment tracking, allowing researchers to reproduce results by snapshotting configurations and data states.

Videbat emphasizes interoperability through standard formats and a plugin architecture. It supports common schemas such as COCO and MOT, along with custom schemas, and offers APIs for automation, scripting, and integration with machine learning frameworks.

Applications span autonomous driving, robotics, surveillance analytics, sports analytics, and educational research. Privacy-preserving features, metadata governance, and access controls are integrated to help organizations manage sensitive footage and comply with data protection requirements.

History and development traces back to an international open-source collaboration launched in the early 2020s. Public releases introduced core ingest and annotation capabilities, followed by features for collaboration, experiment management, and interoperability with external tools.