#8686th of 90 in Open modelsFor image, video and design, 3d
objectron
Dataset of object-centric videos with 3D bounding box and camera pose annotations
- Stars
- 2.4k
- License
- custom license
- Last commit
- Mar 2026
- Last release
- Apr 2021
- Language
- Jupyter Notebook
2,322 on 25 Apr 20262,362 on 14 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
Objectron is a dataset of 15K short object-centric video clips and over 4M annotated images across nine categories (bike, book, bottle, camera, cereal box, chair, cup, laptop, shoe). Each sample has manually annotated 3D bounding boxes plus AR metadata: camera poses, sparse point clouds and planar surfaces. The repo supplies protobuf schemas, tf.record loaders, 3D IoU evaluation and notebooks for TensorFlow, PyTorch and JAX.
Who it is for: Computer vision researchers training or evaluating 3D object detection models
Strengths
- Manual 3D box labels with camera pose, point clouds and surface planes per video
- Covers 9 object categories, collected in 10 countries across five continents
- Notebooks for TensorFlow, PyTorch, 3D IoU evaluation and NeRF training
- Ships tf.Example and tf.SequenceExample shards plus train/test split indexes
Weaknesses
- Full dataset is 4.4TB (1.9TB raw), so downloading needs substantial storage
- Released under C-UDA-1.0, a data-use agreement rather than a standard open-source license
- Only nine fixed object categories
- Last release was April 2021; the README says it is not officially supported
What it needs
- GPU optional
- Needs Google Cloud Storage bucket, TensorFlow, PyTorch, Apache Beam
- Models: tf.Example, tf.SequenceExample, protobuf
What changed
From the release notes. No flag means the note does not say so, not that the upgrade is safe.
Also in Open models
See all 90| Rank | Project | Score |
|---|---|---|
| 1 | ruviewWiFi CSI sensing on ESP32 for presence, breathing, heart rate and pose | 94 out of 100 |
| 2 | layaSingle-pass multilingual classifier returning typed decisions with calibrated odds | 88 out of 100 |
| 3 | rf-detrReal-time transformer models for object detection, segmentation and keypoints | 82 out of 100 |
| #4 | timesfmPretrained time-series foundation model for zero-shot forecasting from Google Research | 80 out of 100 |
| #5 | open_clipPyTorch library for training and running CLIP-style image-text models | 79 out of 100 |