What it does

MyVision is a free, browser-based image annotation tool for building computer-vision training datasets. You draw bounding boxes and polygons, export to formats like COCO, YOLO, and VGG, and optionally let a pre-trained COCO-SSD model auto-label objects for you. Everything stays local—no upload to a server.

The interesting bit

The auto-annotation runs TensorFlow’s COCO-SSD model directly in the browser via TensorFlow.js, which means your images never leave your machine. That’s a genuine privacy win in a category where most tools are SaaS products hungry for your data.

Key highlights

• Supports polygon and bounding-box annotation with point-level editing (add, remove, adjust vertices)
• Exports to COCO JSON, YOLO, VGG, and other common CV formats; can also convert between them by importing and re-exporting
• Auto-labeling with COCO-SSD for quick bounding-box generation
• Zero backend: open index.html and start working, or run npm run watch for development
• UI available in English and Mandarin

Caveats

• The README doesn’t specify how well COCO-SSD handles niche object categories; expect generic classes (person, car, dog, etc.) and manual cleanup for domain-specific work
• 609 stars suggests a modest user base; long-term maintenance is unclear

Verdict

Worth a look if you need a quick, private annotation session without spinning up infrastructure. Skip it if you need team collaboration, versioning, or pixel-perfect semantic segmentation—this is a solo tool, not a platform.