Robust Lifting of Open-World 2D Bounding Boxes with Boxer @FacebookOpenSource
Robust Lifting of Open-World 2D Bounding Boxes with Boxer  @FacebookOpenSource
Uploaded July 2026 | Updated November 2026, 4 hours ago
Daniel DeTone, Research Scientist at Meta Reality Labs Research, presents Boxer at CVPR 2026 in Denver, Colorado. Boxer lifts 2D object detections into static, global, fused 3D oriented bounding boxes (OBBs) from posed images and semi-dense point clouds. It is focused on indoor object detection and comes with a pre-trained model for running inference on a variety of data sources.

🔗 Links & Resources:
Boxer webpage: facebookresearch.github.io/boxer
Boxer paper: arxiv.org/abs/2604.05212

Boxer was trained using NymeriaPlus, a full body pose estimation dataset created with multiple wearable egocentric devices, including Project Aria research glasses.

Learn more about related work, NymeriaPlus: arxiv.org/abs/2603.18496
Robust Lifting of Open-World 2D Bounding Boxes with BoxerProject Aria CVPR 2022 Tutorial: Introducing the Aria Pilot Dataset (2 of 11)The Diff Episode 17: Chatting Velox and Accelerating Data Management with Masha BasmanovaPyrefly v1.0.0 is here!The Diff Episode 15: How Rooms Work in SharedSpacesELI5: idbThe Diff Episode 16: Getting the First Hand Sample and Interaction SDKAria Gen 2 Case Study: Envision - Spatial Audio Navigation and AI Research (with audio descriptions)From Hands to 3D: Egocentric Perception with Aria Glasses with Alain Pagani of DFKIProject Aria CVPR 2023 Tutorial: Introduction to Project Aria (Section 2 of 10)Project Aria CVPR 2023 Tutorial: Aria Digital Twin Dataset (Section 8 of 10)ELI5: React
Meta Open Source |

Robust Lifting of Open-World 2D Bounding Boxes with Boxer

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER