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
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










