An abandoned but functional image-and-video annotation workbench

What it does

VoTT is a React/Redux application—packaged as an Electron desktop app or served as a standalone web app—for manually labeling images and video frames. It connects to local or cloud storage providers (Azure Blob Storage, Bing Image Search, or the local filesystem) to pull in assets, then exports labeled data to formats like TensorFlow (Pascal VOC and TFRecords), Azure Custom Vision, CNTK, CSV, or a generic JSON schema.

The interesting bit

The video labeling workflow shows real attention to tedium: you set a frame-extraction rate, scrub through a timeline color-coded for visited and tagged frames, and jump directly between frames that already have tags. Keyboard hotkeys and tag locking suggest the developers actually spent time labeling things themselves.

Key highlights

• Runs fully offline as an Electron app, or in a modern browser with cloud-only storage
• Extensible provider model for importing from and exporting to local or cloud storage
• Exports to multiple ML formats including TensorFlow, Azure Custom Vision, and CNTK
• Per-project security tokens encrypt sensitive settings so project files can be shared safely
• V2 can convert legacy V1 project JSON files, though V1 itself is frozen

Caveats

• Microsoft has stopped maintaining VoTT; the README explicitly states it is “no longer being maintained”
• The web build cannot access the local filesystem, forcing cloud-based import and export
• V1 lives on a separate branch with no fixes or updates planned

Verdict

A viable choice if you need a no-cost, offline-capable annotator with Azure-friendly export options and can tolerate a frozen codebase. Teams wanting active support should look elsewhere.