A Scalable Pipeline for Making Multi-Task Mid-Level Vision Datasets from 3D Scans
Project Website · Docs · Annotator Repo · Starter Data · >> [Tools] << · Paper Code
This repository includes strong pretrained models for depth and surface normal estimation, training code, dataloaders, starter dataset download and upload utilities, the first publicly available implementation of MiDaS training code, an implementation of the 3D image refocusing augmentation introduced in the paper, and more (detailed in the docs).
Install this package: pip install 'omnidata-tools'
Documentation: https://docs.omnidata.vision for details of this package.
Project Overview: The project website or the ICCV21 paper provide a broad overview of the project.
Citation: If you find the code or models useful, please cite the paper:
@inproceedings{eftekhar2021omnidata,
title={Omnidata: A Scalable Pipeline for Making Multi-Task Mid-Level Vision Datasets From 3D Scans},
author={Eftekhar, Ainaz and Sax, Alexander and Malik, Jitendra and Zamir, Amir},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
pages={10786--10796},
year={2021}
}