Objectron: A Large Scale Dataset of Object-Centric Videos in the Wild with Pose Annotations

3D object detection has recently become popular due to many applications in\nrobotics, augmented reality, autonomy, and image retrieval. We introduce the\nObjectron dataset to advance the state of the art in 3D object detection and\nfoster new research and applications, such as 3D object tracking, view\nsynthesis, and improved 3D shape representation. The dataset contains\nobject-centric short videos with pose annotations for nine categories and\nincludes 4 million annotated images in 14,819 annotated videos. We also propose\na new evaluation metric, 3D Intersection over Union, for 3D object detection.\nWe demonstrate the usefulness of our dataset in 3D object detection tasks by\nproviding baseline models trained on this dataset. Our dataset and evaluation\nsource code are available online at http://www.objectron.dev\n