Skip to main content

Aria Everyday Objects Dataset

Overview​

Aria Everyday Objects (AEO) is a small, challenging 3D object detection dataset for egocentric data. AEO consists of approximately 45 minutes of egocentric data across 25 sequences captured by non-computer vision experts collected in a diverse set of locations throughout the US. Oriented 3D bounding boxes have been annotated for each sequence. Annotation is done in 3D, using the camera calibration, SLAM trajectory and SLAM semi-dense point cloud to assist with annotation. 1037 3D object bounding box instances across 17 classes: Bed, Chair, Couch, Door, Floor, Lamp, Mirror, Plant, Refrigerator, Screen, Sink, Storage, Table, Wall, WallArt, WasherDryer, and Window. The dataset is designed to accelerate research in egocentric 3D perception.

For more details, please visit the project page, view the dataset online, explore the github repository and read the ECCV 2024 paper.

What is in the dataset?​

  • Aria RGB video at 10 Hz
  • Aria SLAM x 2 video at 10 Hz
  • Aria IMU x 2 data
  • High-quality static 3D bounding boxes of 17 object classes

How to download the dataset?​

Visit the AEO webpage here. Scroll down, enter your email, then click the button “Access the Dataset”. Follow the instructions there.

BibTeX Citation​

@article{straub24efm,
      title={EFM3D: A Benchmark for Measuring Progress Towards 3D Egocentric Foundation Models},
      author={Julian Straub and Daniel DeTone and Tianwei Shen and Nan Yang and Chris Sweeney and Richard Newcombe},
      booktitle={arXiv preprint arXiv:2406.10224},
      year={2024},
      url={https://arxiv.org/abs/2406.10224},
}

License​

AEO dataset is released under a non-commercial license described here.

Contributors​

Julian Straub, Daniel DeTone, Tianwei Shen, Nan Yang, Chris Sweeney, Richard Newcombe