Best data annotation tools that are open source
Best data annotation tools, if you want open source: CVAT for boxes on images, video and 3D point clouds, Label Studio for mixed data types, and doccano or INCEpTION for text. The table compares their license, data types, export formats, how each is self-hosted and whether its repository is archived, with a source link on every fact.
Data annotation tools comparison
Pick by your data first: most tools handle only images, or only text. Then check the export format your training code reads.
Download the table as CSV. Source: each project's own repository and documentation, linked in each row. License CC BY 4.0: copy, edit or publish the list, and credit dataannotationservices.io with a link.
Open-source tools by kind of data
Images and video. CVAT covers boxes, polygons and tracking across video frames, and runs on your own server with Docker. For one person labeling photos on a laptop, labelme, X-AnyLabeling or make-sense in the browser are lighter. LabelImg still works, but its repository is archived.
Text. doccano is the quickest to start for tagging names and classifying text; INCEpTION suits research teams that need custom layers and CoNLL-U. Argilla is built for rating model outputs, so it fits chatbot feedback work.
Audio. Label Studio is the one tool in the table that labels audio next to images and text. See audio annotation services if you want people to do it.
3D point clouds. CVAT and X-AnyLabeling both list point-cloud labeling. For large driving datasets, see LiDAR annotation.
For the head-to-head of the two most used tools, read CVAT vs Label Studio; to move exports between tools, use the bounding box format converter. Paid platforms such as Labelbox and Encord are in the data labeling platform comparison. If you would rather have people do the labeling, send one brief to three annotation companies.
How this table was made
Open source here means an OSI license on the project's main repository. Each row was read from the project's own repository and documentation as of the release named in that row, so newer releases may differ.