NVlabs/BundleSDF

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

1410 stars · 169 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

BundleSDF is a CVPR 2023 research implementation from NVIDIA for near real-time 6-DoF pose tracking of unknown rigid objects from monocular RGBD video, jointly performing neural 3D reconstruction via a Neural Object Field. It requires only a first-frame segmentation mask and combines feature matching, pose graph optimization, and a dynamic memory pool of posed frames.

Use cases

• track 6-DoF pose of an unknown object from RGBD video
• reconstruct 3D geometry and appearance of untextured objects
• run object tracking under occlusion and large pose changes
• benchmark on HO3D, YCBInEOAT, and BEHAVE datasets
• generate neural object fields for robotics manipulation research

When to choose

• you need to track arbitrary rigid objects without CAD models or prior training
• you have monocular RGBD input and only a first-frame segmentation
• you want state-of-the-art joint pose tracking and neural 3D reconstruction
• you are doing robotics or computer-vision research on object pose estimation

When to avoid

• you need a production-ready supported product rather than research code
• you lack a GPU or cannot use the Docker environment
• you have no RGBD data and only monocular RGB
• you need real-time performance on embedded hardware

Facets

library · maturity stable

computer-vision machine-learning graphics simulation computer-vision robotics machine-learning python 6dof-pose-tracking neural-object-field rgbd 3d-reconstruction slam cvpr-2023 research-code linux docker gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="NVlabs/BundleSDF")