Google announced Objectron on November 9, 2020, as a dataset for research into 3D object understanding. It pairs short videos that circle everyday objects with 3D bounding-box annotations and augmented-reality (AR) session data. Google also released companion object-detection models through MediaPipe, but the announcement describes research resources and potential applications—not proof that Objectron outperforms other datasets or systems.
What is the Objectron dataset?
Objectron is a collection of object-centered video clips rather than a conventional set of isolated photographs. In each clip, a camera moves around an everyday object to capture it from different views. Google Research published the announcement on November 9, 2020; its authors were Google Research software engineers Adel Ahmadyan and Liangkai Zhang. The post framed the dataset as a response to a shortage of large real-world datasets for 3D object understanding compared with photo-based 2D computer vision.
The idea is to give researchers video and camera-stream material that shows more of an object’s 3D structure. Google identified augmented reality, robotics, autonomy, and image retrieval as possible application areas. Those were motivations for the work, not demonstrated outcomes attributable to the dataset.
How much data does Objectron include?
Google Research reported 15,000 annotated video clips and more than 4 million annotated images, collected across 10 countries on five continents. These are Google-published counts, not independently audited totals. The Google Research Datasets repository, accessed in 2026, describes the rounded collection as about 15,000 clips and 4 million images and provides per-category clip and frame counts.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitches#1 Best Overall
The repository also reports 1.9 TB for raw videos and annotations and 4.4 TB for the total dataset package, which includes records and sequences. These are repository storage figures; the total depends on the packaged contents and should not be treated as a fixed download size for every access method.
Which objects and annotations are included?
Dataset categories
The repository lists nine categories: bikes, books, bottles, cameras, cereal boxes, chairs, cups, laptops, and shoes. This category list is distinct from the four categories named for the companion detection models: shoes, chairs, mugs, and cameras. In particular, the dataset calls its category “cups,” while the model announcement says “mugs.”
Video, AR metadata, and 3D boxes
Each clip is accompanied by AR session metadata. Google names camera poses and sparse point clouds; the repository also describes planes in the surrounding environment. Manually annotated 3D bounding boxes specify an object’s position, orientation, and dimensions. Together, the moving-camera footage and geometric annotations let researchers work with object views and spatial information, rather than only 2D image labels.
What did Google release with the dataset?
Google also announced 3D object-detection models trained using Objectron for shoes, chairs, mugs, and cameras. They were released through MediaPipe, Google’s open-source framework for cross-platform machine-learning solutions for live and streaming media. Google AI Edge describes the Objectron pipeline as real-time 3D object detection for mobile devices; see the MediaPipe Objectron documentation.
Recommended Free Tools
The repository includes tutorials for downloading data, loading it with TensorFlow or PyTorch, parsing raw annotations and AR metadata, evaluating with 3D intersection-over-union (IoU), working with sequences, and training NeRF models. Repository release notes also refer to downloadable detection models and Python and Web API examples. These are repository-described resources; their current availability and compatibility may change.
What can researchers use Objectron to study?
The Objectron paper, “Objectron: A Large Scale Dataset of Object-Centric Videos in the Wild With Pose Annotations,” appears in the CVPR 2021 proceedings. Its record describes 3D object detection as the dataset’s central research aim and names 3D object tracking, view synthesis, and improved 3D shape representation as possible research applications. The paper’s inclusion and stated aims do not by themselves establish a particular downstream impact or a performance advantage over another dataset or system.
The 2020 announcement mentions evaluating detection models with 3D IoU, but it does not provide a numerical performance result in the evidence cited here. No head-to-head dataset comparison is established by the cited announcement, repository, paper record, or MediaPipe documentation. Claims that Objectron is larger, more accurate, more diverse, or superior to alternatives therefore require separate comparative evidence.
Can you use Objectron with MediaPipe?
Yes. The release paired the dataset with MediaPipe object-detection models, and MediaPipe documentation describes the pipeline for real-time 3D detection on mobile devices. The dataset itself is also presented as material for research and model development; its tutorials cover data handling and evaluation workflows. The repository names the dataset license as the Computational Use of Data Agreement 1.0 (C-UDA-1.0). Review the license text for its terms before using the data; the license name alone does not establish particular permissions or restrictions.
Free tools Windows power users keep installed
One-click scans. No signup required.
Best Value
Sources and current availability
- Google Research’s November 9, 2020 announcement reports the motivation, collection figures, categories, annotations, and companion models.
- The official Objectron repository describes the dataset contents, per-category counts, storage figures, tutorials, and license name.
- The CVPR 2021 paper record identifies the publication and stated research aims.
- Repository release notes describe release assets and API examples.
Objectron’s announcement and paper are historical releases. Repository files, downloads, model assets, packaging, and software dependencies can change; current download availability and compatibility are not established by those descriptions.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
