The MatrixCity dataset is designed to aid research in city-scale neural rendering. This field has potential real-world applications such as aerial surveying, virtual reality, film production, and gaming, but research has been limited due to the lack of a comprehensive dataset. Collecting real-world data is costly and technically challenging, and cannot offer control over environmental variables. MatrixCity, built using Unreal Engine 5, remedies this by providing a large-scale, high-quality synthetic dataset that includes both aerial and street views of two city maps. The dataset, which will be publicly available, also includes additional data modalities and flexible controls over environmental factors.
Publication date: 28 Sep 2023
Project Page: https://city-super.github.io/matrixcity/
Paper: https://arxiv.org/pdf/2309.16553
