An evaluation of DUSt3R/MASt3R/VGGT 3D reconstruction on photogrammetric aerial blocks

Abstract

State-of-the-art 3D computer vision algorithms continue to improve on sparse, unordered image sets. Recently developed foundational models for 3D reconstruction, such as dense and unconstrained stereo 3d reconstruction (DUSt3R), matching and stereo 3d reconstruction (MASt3R), and visual geometry grounded transformer (VGGT), have attracted considerable attention due to their ability to handle very sparse image overlaps, as well as their generalization capability. In light of this contribution, evaluating DUSt3R/MASt3R/VGGT on typical aerial images is important, as these models may hold the potential to handle extremely low image overlaps, stereo occlusions, and textureless regions. For highly redundant collections, they can accelerate 3D reconstruction by using extremely sparsified image sets. Despite being tested on various computer vision benchmarks, their potential on photogrammetric aerial blocks remains unexplored. We present a comprehensive evaluation of the pre-trained DUSt3R/MASt3R/VGGT models on the aerial blocks of the UseGeo dataset for pose estimation and dense 3D reconstruction. The methods reconstruct dense point clouds from very sparse inputs (fewer than ten images, resized to a maximum dimension of 518 pixels), achieving reasonable accuracy and completeness gains up to 50% over COLMAP. VGGT further shows superior computational efficiency, scalability, and more reliable camera pose estimation. However, all three show limitations on high-resolution imagery and large image sets, with the camera pose estimation reliability significantly declining as the number of images and the geometric complexity of the scene increase. These findings indicate that while transformer-based methods cannot replace traditional SfM and MVS methods entirely, they hold potential as complementary approaches, especially in challenging, low-resolution, and extremely sparse scenarios.