Funes Releases Open 3D Architecture Dataset for Radiance Fields

Going out to do captures for radiance field reconstructions can sometimes feel intimidating and hard to figure out where to start. It can also be hard to find available sample raw datasets of images that you can download and practice with. That’s why we’re excited to spotlight a new open dataset released by the team at Funes, designed specifically to support experimentation and benchmarking in architectural 3D reconstruction.

Funes is a digital museum dedicated to preserving and curating 3D representations of human made structures. They’re building a global archive of the built environment, and always looking for new contributors.

The Funes 3D Architecture Dataset contains over 10,000 high resolution photos of 20 distinct buildings, each captured with consistency and clarity in mind. The image sets were all shot over short time windows, so you won’t have to deal with shifting shadows, changing skies, or inconsistent lighting. 

The entirety of the dataset comes in at about 75 GB and retains both the images in JPG, the associated metadata, and even comes with pre-processed COLMAP data. This dataset is a great fit for testing structure from motion (SfM), or even jumping straight into Gaussian Splatting training runs. It’s also an ideal playground for newcomers who want to try their hand at running their first reconstruction without needing access to expensive scanning gear.

We’re always excited to see new contributions that make radiance field research more approachable and reproducible. As the tooling around Radiance Fields continues to mature, having access to high quality training data becomes just as important as the rendering techniques themselves. Big thanks to Funes for opening up their archives and helping move the community forward.

The dataset is offered free of charge for academic and educational purposes under a CC BY-NC 4.0 license. That means you can download it, train with it, experiment, and publish, just no commercial use. Funes kindly asks for attribution in any resulting papers, posts, or presentations.

To request access, fill out the short form on the Funes website.