Objectron contains 15,000 object-centric video clips and 4 million annotated images across nine categories, produced by Google Research and updated in 2026. Each video includes AR session metadata such as camera poses, sparse point-clouds, and planes, with objects annotated via 3D bounding boxes.
Use Cases
- Training 3D object detection models using the 3D bounding box coordinates for position and orientation
- Developing 6D pose estimation algorithms using the provided camera poses and AR session metadata
- Evaluating 3D reconstruction techniques using the sparse point-clouds and plane data
Strengths
- 4 million annotated images across 15,000 video clips
- Includes AR session metadata like camera poses and point clouds
- 3D bounding boxes providing position, orientation, and dimensions
Limitations
- Limited to nine specific object categories
- Focuses on single-object centric clips rather than complex multi-object scenes
Provenance
- Source
- google-research-datasets
- Collection Method
- Video capture with AR session recording and 3D bounding box annotation
- Freshness
- Last updated March 2026