render-tag Documentation

Welcome to the official documentation for render-tag, a high-performance procedural 3D synthetic data generator for fiducial marker training.

render-tag is designed to bridge the gap between photorealistic 3D rendering and high-precision computer vision requirements, specifically for AprilTag and ArUco detection.

Core Documentation

• User Guide: Installation, CLI usage, and configuration presets.
• Architecture: Deep dive into the Host-Backend design and "Hot Loop" implementation.
• Coordinate Systems & Standards: Canonical geometric contracts, pose conventions, and output data formats.
• Benchmarking & Auditing: Tracking performance and verifying dataset quality.
• API Reference: Low-level Python API documentation.

Key Features

• Procedural Scene Generation: Deterministic generation of complex 3D scenes with randomized lighting, textures, and physics.
• Host-Backend Architecture: Decouples heavy 3D rendering (Blender) from generation logic (Python), enabling high-throughput pipelines.
• Sub-pixel Accuracy: Optimized Cycles rendering configurations ensuring edge and corner integrity.
• Rich Annotations: Comprehensive ground truth including 6DoF poses, PPM, and visibility metrics.
• Hugging Face Integration: Native support for managing assets and datasets on the Hub.