red

A 3D multi-camera labeling tool for fast review and triangulation across many synchronized video streams.

Overview

red is the labeling counterpart to orange. It takes multi-view video (typically recorded with orange) and lets you label keypoints across all camera views simultaneously, with real-time GPU decoding (h264 / hevc), synchronized playback across all cameras, and multi-view triangulation. Labeled data can be exported for downstream training (YOLO detection, YOLO pose, JARVIS).

Source on GitHub

Video demo

Features

1. Real-time GPU-accelerated decoding (h264, hevc)
2. Synchronized playback across multiple cameras
3. Multi-view keypoint labeling and triangulation
4. Skeleton authoring with skeleton.json
5. Exporters for YOLO and JARVIS training pipelines

Where to start

• New install: Installation — dependencies and build.
• Already built: Configuration (default folders, project layout).
• Exporting labels for training: Data export.