Welcome to Lab4D’s documentation!

Lab4D is a framework for 4D reconstruction from monocular videos.

Features

• Representation

  • neural implicit representation

  • deformation fields (neural fields, control-points, skeleton)

  • compositional scene

  • category-level models

• Interface for priors

  • pixelwise priors: depth, flow, DINOv2 features

  • segmentation: track-anything, video instance segmentation

  • camera viewpoint: viewpoint network, manual annotation

• Efficiency

  • multi-gpu training

  • dual-quaternion ops

Note

This is an alpha release and the APIs are subject to change as we continuously improve and refine it. We encourage users to provide feedback and report bugs via github issues. Thank you for your support.