What it does ImgLab is a purely browser-based image annotation tool for drawing bounding boxes, polygons, and landmark points on images. It exports to multiple ML formats—dlib XML/pts, Pascal VOC, COCO—and stores project files in a compact custom format called Nimn so they can be emailed around. No server, no install, minimal CPU and memory.

The interesting bit The dlib-specific quality-of-life features are the unusual angle: you can drag landmark labels up and down to reorder feature points after creating them, rather than drawing them in the exact right sequence from the start. There’s also a Face++ plugin for auto-suggesting landmarks, which saves some tedious clicking if you’re doing face work.

Key highlights

• Runs entirely in the browser; zero backend or dependencies
• Exports to dlib XML/pts, Pascal VOC, COCO (TensorFlow support is “in plan”)
• Reorder landmark points by dragging labels instead of redrawing
• Autosaves to browser cache; exports to disk when you’re ready
• Keyboard shortcuts, zoom, tracking lines, and adjustable image opacity for precision work

Caveats

• The project is actively looking for maintainers and contributors, which suggests development may be slow or stalled
• “Tensorflow (in plan)” appears to have been in plan for some time; don’t count on it
• Ellipses, lines, and curves are listed as future shapes “if there is demand”

Verdict Good fit if you need a lightweight, no-install annotator for dlib or COCO workflows and don’t mind a slightly dated UI. Skip it if you need team collaboration, version control, or guaranteed active maintenance—those aren’t here.