A browser-based labeller that never phones home

What it does

makesense.ai is a browser-based photo labelling tool built with TypeScript and React/Redux. You draw bounding boxes, polygons, points, lines, and labels, then export them to formats like YOLO, COCO JSON, Pascal VOC XML, or CSV. It targets small computer-vision projects and requires no server-side image processing.

The interesting bit

The AI integrations run entirely in the browser via TensorFlow.js, so your photos never leave the device. It bundles YOLOv5 (via yolov5js), SSD, and PoseNet to suggest annotations or automate repetitive labelling, which is a rare privacy-first approach for a free tool.

Key highlights

• Runs fully in the browser; no installation or cross-platform compatibility issues.
• Supports multiple export formats depending on annotation type: YOLO and VOC XML for rectangles, COCO JSON and VGG JSON for polygons, CSV for points and lines.
• Can load custom YOLOv5 models exported to TensorFlow.js format, not just pretrained zoo models.
• Keyboard-driven workflow with shortcuts for image navigation, zoom, and label selection.
• GPL-3.0 licensed.

Caveats

• Documentation is newly released but still under construction, so some topics may be missing.
• Local development requires specific Node.js and npm versions (v16.x.x and 8.x.x), and the README notes related issues.
• Import support is narrower than export: only rectangles and polygons can be imported, and from a limited set of formats.

Verdict

Worth a look if you need a quick, private way to annotate a small image dataset for object detection without spinning up a backend. Skip it if you need mature documentation or broad import support for points, lines, and polygons.