DUSt3R: Geometric 3D Vision Made Easy
By Shuzhe Wang, Vincent Leroy, Yohann Cabon, Boris Chidlovskii, Jérôme Revaud. Computer Vision and Pattern Recognition, 2023
1,664 citations recorded by the source corpus.
Multi-view stereo reconstruction (MVS) in the wild re-quires to first estimate the camera intrinsic and extrinsic parameters. These are usually tedious and cumbersome to obtain, yet they are mandatory to triangulate corresponding pixels in 3D space, which is at the core of all best performing MVS algorithms. In this work, we take an opposite stance and introduce DUSt3R, a radically novel paradigm for Dense and Unconstrained Stereo 3D Reconstruction of arbitrary image collections, operating without prior infor-mation about camera calibration nor viewpoint poses. We cast the pairwise reconstruction problem as a regression of pointmaps, relaxing the hard constraints of usual projective camera models. We show that this formulation smoothly unifies the monocular and binocular reconstruction cases. In the case where more than two images are provided, we fur-ther propose a simple yet effective global alignment strategy that expresses all pairwise pointmaps in a common refer-ence frame. We base our network architecture on standard Transformer encoders and decoders, allowing us to leverage powerful pretrained models. Our formulation directly provides a 3D model of the scene as well as depth information, but interestingly, we can seamlessly recover from it, pixel matches, focal lengths, relative and absolute cameras. Extensive experiments on all these tasks showcase how DUSt3R effectively unifies various 3D vision tasks, setting new performance records on monocular & multi-view depth estimation as well as relative pose estimation. In summary, DUSt3R makes many geometric 3D vision tasks easy. Code and mod-els at https://github.com/naver/dust3r.
Subfields: Geology
View Paper: https://www.semanticscholar.org/paper/5f82a81766cb78395a55b8fc697c2421a20f4a9e
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Geology
- Ming-Hsuan Yang
- Trevor Darrell
- Alexei A. Efros
- Vincent Leroy
- Angjoo Kanazawa
- Aleksander Holynski
- Nanxuan Zhao
- Kalyan Sunkavalli
Citations in the corpus
- GS-LRM: Large Reconstruction Model for 3D Gaussian Splatting (2024, background): “To make our model more practical, it is interesting to explore ways to get rid of the camera parameters from the input end [28,63,65].”
- Fast3R: Towards 3D Reconstruction of 1000+ Images in One Forward Pass (2025, positive): “This traditional “pipeline” paradigm has recently been challenged by DUSt3R [61], which directly predicts 3D structure from RGB images.”
- Continuous 3D Perception Model with Persistent State (2025, positive): “These two stages are trained on 224 × 224 images to reduce computational costs, following DUSt3R [86].”
- Gen3C: 3D-Informed World-Consistent Video Generation with Precise Camera Control (2025, neutral): “Recently, depth estimation has achieved significant progress across various domains [20, 24, 56, 64, 65], such as indoor, outdoor, or self-driving scenes.”
- MonST3R: A Simple Approach for Estimating Geometry in the Presence of Motion (2024, methodology): “Our architecture is based on DUSt3R (Wang et al., 2024c), a ViT-based architecture (Dosovitskiy et al., 2021) that is pre-trained on a cross-view completion task (Weinzaepfel et al., 2023) in a self-supervised manner.”
- VGGT: Visual Geometry Grounded Transformer (2025, neutral): “Importantly, like in DUSt3R [87], the point maps are viewpoint invariant , meaning that the 3D points P i ( y ) are defined in the coordinate system of the first camera g 1 , which we take as the world reference frame.”
- Depth Anything 3: Recovering the Visual Space from Any Views (2025, neutral): “Note that using Scannet++ for training is fair to other methods, as it is widely used for training in [91, 96].”
- Grounding Image Matching in 3D with MASt3R (2024, methodology): “Finally, the recent DUSt3R [102] explore the possibility of recovering correspondences from the a-priori harder task of 3D reconstruction from uncalibrated images.”
- ViewCrafter: Taming Video Diffusion Models for High-fidelity Novel View Synthesis (2024, methodology): “Although InstantSplat [79] utilizes DUSt3R [19] for point cloud initialization, which better preserves details from the ground truth training images, it fails to recover occlusion regions due to its omission of the densification”