OpenBox: Annotate Any Bounding Boxes in 3D

Name: OpenBox: Annotate Any Bounding Boxes in 3D
Uploaded: 2026-10-06T12:04:50.025782+00:00
Duration: 344 s
Description: OpenBox creates an automatic pipeline to generate high-quality 3D bounding boxes for vehicles, pedestrians, and cyclists using 2D vision models without self-training. It works by aligning 2D image cues with LiDAR point clouds and refining these points based on context. The system then categorizes objects by physical state—static, dynamic rigid, or deformable—to produce adaptive 3D boxes.

OpenBox creates an automatic pipeline to generate high-quality 3D bounding boxes for vehicles, pedestrians, and cyclists using 2D vision models without self-training. It works by aligning 2D image cues with LiDAR point clouds and refining these points based on context. The system then categorizes objects by physical state—static, dynamic rigid, or deformable—to produce adaptive 3D boxes.

Open the full episode