Heritage3DMtl: An Open Multi-Modal UAV Dataset of Montreal’s Diverse Heritage Buildings for Research

Abstract

This work introduces Heritage3DMtl, a real-world, multi-view UAV dataset comprising 17 architecturally diverse heritage buildings in Montreal. The dataset includes images, Structure-from-Motion (SfM) outputs, point clouds, and triangle meshes reconstructed using open-source photogrammetry tools. It is intended as an openly available dataset for research in 3D heritage reconstruction, computer vision, neural rendering, architectural analysis, and digital heritage preservation. The paper describes a practical acquisition workflow based on low-cost consumer-grade UAVs together with the associated processing pipeline. UAV-based acquisition was selected because it provides flexible viewpoints around complex heritage structures that are often difficult to obtain using ground-based photography or publicly available airborne imagery alone. The resulting dataset supports multiple downstream applications, including LOD3 reconstruction, Neural Radiance Fields (NeRFs), and 3D Gaussian Splatting for novel-view synthesis. Compared with many publicly documented heritage datasets that focus on one or a few monuments, Heritage3DMtl comprises 17 heritage buildings representing a broad range of architectural styles across the Montreal region. The dataset is publicly available through Zenodo at https://zenodo.org/records/13370334.