A labeling tool that admits paperwork is still paperwork

What it does

LOST (Label Objects and Save Time) is a browser-based system for organizing collaborative image annotation. You set up pipelines—either pre-built ones for bounding boxes, polygons, points, and lines, or custom ones you stitch together—then distribute the work to annotators anywhere. It connects to external storage like S3 or Azure Blob, tracks progress with statistics, and exports datasets on demand.

The interesting bit

The semi-automatic angle: LOST can inject AI-generated annotation proposals into the human workflow, so annotators review and correct rather than draw from scratch. The framework also exposes a Jupyter-Lab integration for developing custom pipelines without leaving the environment.

Key highlights

• Pre-built annotation tools for single images (SIA) and image clusters (MIA)
• Pipeline import/export and sharing between colleagues
• LDAP integration and email notifications for enterprise deployments
• Docker-based quick setup with releases on DockerHub
• Scalable design intended to distribute heavy compute across multiple machines

Caveats

• Documentation is still being updated for version 3.x; the README explicitly notes this is “currently still in progress”
• Quick setup steps are only tested for Ubuntu; your mileage may vary elsewhere

Verdict

Worth a look if you need a self-hosted, team-scale annotation platform with some automation hooks. Probably overkill for one-off personal projects or teams already embedded in commercial tools like Labelbox or CVAT.