AI Computer Vision & Graphics Research

Introducing Holicity: A City-Scale Dataset for Learning Holistic 3D Structures

HoliCity, a city-scale dataset and all-in-one data platform for research into learning abstracted high-level holistic 3D structures derived from city CAD (computer-aided design) models.

Researchers from UC Berkeley, Stanford and Bytedance recently introduced HoliCity, a city-scale dataset and all-in-one data platform for research into learning abstracted high-level holistic 3D structures derived from city CAD (computer-aided design) models. A rigorous annotation pipeline and associated tools enable continuous increases in the scale and richness of the dataset, supporting the study and evaluation of a wide spectrum of 3D vision methods, from low-level to high-level.

Screen Shot 2020-08-10 at 2.54.17 PM.png

3D vision technologies have been in increasing demand in recent years. Although the development of robust point features such as SIFT, ORB, SfM and SLAM has greatly improved 3D vision performance; their resulting point clouds are still noisy, incomplete, and difficult to use directly in many real-world applications.

Inspired by the tremendous success of deep convolutional neural networks (CNNs) in image classification, 3D vision researchers have proposed a variety of neural network-based approaches to extract high-level holistic structures from images, such as wireframes, planes, cuboids, vanishing points, room layouts, and building layouts.

HoliCity includes 6,300 high-resolution real-world panoramas accurately aligned with a 3D CAD model of downtown London with more than 20 km^2 of area, and was designed to address the high cost of data preparation. Rather than relying on expensive vehicle-mounted LiDAR scanners, HoliCity leverages existing high-quality 3D CAD city models and street-view imagery from geographic information system (GIS) collection.

The researchers conducted multiple experiments with their CAD model-based data platform for 3D vision research, testing potential applications and the platform’s generalizability from and to other datasets. The results confirm HoliCity’s usefulness as an urban environment data platform for a range of 3D vision research areas.

Currently, HoliCity focuses on modelling building architectures. In the future, the researchers plan to extend the datasets to allow the modelling of moving objects such as pedestrians and automobiles in urban environments.

The paper HoliCity: A City-Scale Data Platform for Learning Holistic 3D Structures is on arXiv.


Analyst: Grace Duan | Editor: Michael Sarazen; Fangyu Cai



Synced Report | A Survey of China’s Artificial Intelligence Solutions in Response to the COVID-19 Pandemic — 87 Case Studies from 700+ AI Vendors

This report offers a look at how China has leveraged artificial intelligence technologies in the battle against COVID-19. It is also available on Amazon KindleAlong with this report, we also introduced a database covering additional 1428 artificial intelligence solutions from 12 pandemic scenarios.

Click here to find more reports from us.


We know you don’t want to miss any story. Subscribe to our popular Synced Global AI Weekly to get weekly AI updates.

0 comments on “Introducing Holicity: A City-Scale Dataset for Learning Holistic 3D Structures

Leave a Reply

Your email address will not be published. Required fields are marked *

%d bloggers like this: