tech

January 22, 2026

D4RT: Unified, Fast 4D Scene Reconstruction & Tracking

Meet D4RT, a unified AI model for 4D scene reconstruction and tracking.

D4RT: Unified, Fast 4D Scene Reconstruction & Tracking

TL;DR

  • D4RT is a unified AI model for dynamic 4D scene reconstruction and tracking.
  • It uses a query-based approach within a Transformer architecture to recover 3D information from 2D video.
  • D4RT can perform point tracking, point cloud reconstruction, and camera pose estimation.
  • The model is significantly more efficient, up to 300x faster than previous methods, enabling real-time applications.
  • D4RT demonstrates superior accuracy and fidelity in handling dynamic objects and complex scenes compared to existing state-of-the-art.
  • Potential downstream applications include robotics, augmented reality, and the development of AI 'world models'.