LabelU is an open-source platform for annotating image, video and audio data, intended for work supporting complex data analysis and model training. Image tools include bounding boxes, semantic segmentation, polylines and keypoints; video and audio tools support segmentation, classification and information extraction. Users can load pre-annotated data for review and adjustment, or use AI model services to detect and segment image objects, including batch annotation with progress tracking. LabelU imports annotation data from S3-compatible storage such as AWS S3 and MinIO, and exports JSON, COCO and MASK formats. Its local setup uses Miniconda, Python 3.11 and pip, with a server at http://localhost:8000/. The platform includes SQLite and supports MySQL installation and migration. Its model server provides HTTP POST / and GET /health endpoints. Listed reference models require about 4GB VRAM for Florence-2 and GroundingDINO with SAM ViT-B, while SAM 3 requires about 8GB and CUDA 12.6 or later. The project is licensed under Apache 2.0.
Who it is for
LabelU suits teams annotating image, video or audio data for analysis and model training. It is aimed at users who can work with a local deployment and, for model-assisted labeling, the listed hardware requirements.
What is good
- Supports image, video and audio annotation.
- Can refine pre-annotated data and use AI model services.
- Imports from S3-compatible storage and exports three formats.
- Apache 2.0 licensed.
What to know first
- Local setup requires Miniconda, Python 3.11 and pip.
- SAM 3 requires about 8GB VRAM and CUDA 12.6 or later.
- The listed support channel is an OpenDataLab WeChat group.
Verdict
LabelU brings multimodal annotation, model-assisted labeling and several export formats into an open-source package. Its local setup and model requirements are relevant considerations for teams planning deployment.
Compared on AI data labeling tools
- Supported modalities
- image, video, audio
- Model-assisted labeling
- Yes
- Human review workflows
- Yes
- Custom ontologies
- Yes
- Deployment options
- self hosted
- API access
- Yes
