AI data platform
LinkedAI
Annotation tools on HTML Canvas, with SAM-assisted segmentation
Annotation, segmentation and review tools for computer vision datasets. I built them on HTML Canvas and integrated Segment Anything Model (SAM) into the annotation workflows.
Context
LinkedAI is an international SaaS platform for preparing the image datasets that train computer vision models. The quality of those datasets depends on the tools people use to draw, correct and review labels.
Problem
Annotators draw and edit regions over images all day, across large collections. A generic interface is too imprecise for that and too slow once state and data volumes grow. And tracing every outline by hand is slow work in itself.
What I owned
- Image annotation, segmentation and review tools on HTML Canvas: precise editing interactions, and the complex state behind them.
- The features to organize projects and image collections and to review large volumes of data, for responsiveness and for clearer review flows.
- Integrating Segment Anything Model (SAM) into the annotation workflows, so AI-assisted segmentation is available when creating and editing labels.
- My part of SAM was the integration into the tools. I did not research or train the model.
Decisions
- Canvas rather than the DOM for the annotation surface, to keep drawing and editing precise and responsive over large images.
- Keep the state of projects, image collections, annotations and review separate, so each could change without dragging the others.
- Put SAM-assisted segmentation inside the same label creation and editing flow, not in a separate tool, so the annotator keeps working in one place.
Result
Canvas-based annotation, segmentation and review tools with SAM-assisted segmentation in the annotation workflows. The platform is no longer publicly available, so there is no live link.