EVL: Egocentric Voxel Lifting

[paper]

Intro

Egocentric Voxel Lifting (EVL) is a baseline model for 3D object detection and surface reconstruction. EVL is trained on a large simulation dataset Aria Synthetic Enviroments. EVL leverages egocentric modalities and inherits foundational capabilities from 2D foundation models.

Usage

Please refer to the EFM3D repo for installation and how to use the model

Citing EVL

@article{straub2024efm3d,
  title={EFM3D: A Benchmark for Measuring Progress Towards 3D Egocentric Foundation Models},
  author={Straub, Julian and DeTone, Daniel and Shen, Tianwei and Yang, Nan and Sweeney, Chris and Newcombe, Richard},
  journal={arXiv preprint arXiv:2406.10224},
  year={2024}
}

License

EVL is released by Meta under the Apache 2.0 license.