A small sample dataset from the Omnidata Starter Dataset in the following paper:
A Scalable Pipeline for Making Multi-Task Mid-Level Vision Datasets from 3D Scans
This is only one sample scene in our GSO + Replica dataset split, which is created by scattering Google Scanned Objects around Replica buildings using the Habitat environment. The full Omnidata Starter Dataset contains more than 14 million images from over 2000 spaces.
For more details and the full dataset please see the main dataset repository and the project website.