RHINO reconstructs the human, the manipulated object, and the scene in 4D within a common world frame from a moving monocular camera.

GauSTAR binds Gaussians to mesh faces to achieve photorealistic rendering, accurate surface reconstruction, and reliable 3D tracking of dynamic scenes, even when their topology changes.

ODHSR jointly tracks the camera and human pose and reconstructs both the human and the scene with 3D Gaussians, online from a single monocular video.

HSR jointly reconstructs the dynamic human and the static scene as neural implicit surfaces from a monocular RGB video.

A hybrid representation that stores local geometry and texture features on the vertices of a skinned body model, enabling neural avatars that are fully editable and customizable.

Makes COLMAP robust to repeated and symmetric structures by filtering out wrong matches with two disambiguation methods.

A CUDA fixed-radius nearest neighbor search with a pytorch3d.ops.knn_points-style interface, over 10× faster than existing GPU KNN implementations.