A Scalable Pipeline for Making Multi-Task Mid-Level Vision Datasets from 3D Scans
Annotator Repo
· >> [Starter Data] <<
· Tooling + Training Repo
· Reference Code
· Project Website
This repository contains the Starter Dataset generated by the Omnidata Annotator from our paper:
Omnidata: A Scalable Pipeline for Making Multi-Task Mid-Level Vision Datasets from 3D Scans (ICCV2021)
We provide a Starter Dataset generated by Omnidata Pipeline from some existing 3D datasets. It contains more than 14 million
images from over 2000 spaces
with 21 different mid-level vision cues
per image. The dataset covers very diverse scenes (indoors and outdoors) and views (scene- and object-centric).
The full starter dataset will be available to download soon.
We provide a sample data from a random building in our GSO + Replica dataset split, which is created by scattering Google Scanned Objects around Replica buildings using the Habitat environment. This is only a sample scene (with mostly object-centric views) from over 2000 scenes
available in the full dataset.
You can download and untar the sample data with the following command:
wget https://drive.switch.ch/index.php/s/MkygxW0WLiLKsNz/download
tar -xf download
Now the sample dataset is available in the folder omnidata_sample_dataset
.
Sample Data (GSO+Replica) |
---|
![]() |
Taskonomy | Replica | GSO+Replica | HM3D | |
---|---|---|---|---|
Field of View | ![]() |
![]() |
![]() |
![]() |
Camera Pitch | ![]() |
![]() |
![]() |
![]() |
Camera Roll | ![]() |
![]() |
![]() |
![]() |
Obliqueness Angle | ![]() |
![]() |
![]() |
![]() |
Camera Distance | ![]() |
![]() |
![]() |
![]() |
Views per Point | ![]() |
![]() |
![]() |
![]() |
If you find this dataset useful in your research, please cite our 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}
}