Easi3R: Estimating Disentangled Motion from DUSt3R Without Training
About
Recent advances in DUSt3R have enabled robust estimation of dense point clouds and camera parameters of static scenes, leveraging Transformer network architectures and direct supervision on large-scale 3D datasets. In contrast, the limited scale and diversity of available 4D datasets present a major bottleneck for training a highly generalizable 4D model. This constraint has driven conventional 4D methods to fine-tune 3D models on scalable dynamic video data with additional geometric priors such as optical flow and depths. In this work, we take an opposite path and introduce Easi3R, a simple yet efficient training-free method for 4D reconstruction. Our approach applies attention adaptation during inference, eliminating the need for from-scratch pre-training or network fine-tuning. We find that the attention layers in DUSt3R inherently encode rich information about camera and object motion. By carefully disentangling these attention maps, we achieve accurate dynamic region segmentation, camera pose estimation, and 4D dense point map reconstruction. Extensive experiments on real-world dynamic videos demonstrate that our lightweight attention adaptation significantly outperforms previous state-of-the-art methods that are trained or finetuned on extensive dynamic datasets. Our code is publicly available for research purpose at https://easi3r.github.io/
Key insights
What problem does Easi3R solve and how?
DUSt3R reconstructs static scenes but fails on dynamic videos. Easi3R adapts it training-free by disentangling camera and object motion from cross-attention maps, enabling dynamic segmentation, pose estimation, and 4D reconstruction via a second inference pass.
How does it differ from prior dynamic 4D methods?
Unlike MonST3R, DAS3R, and CUT3R, which fine-tune on dynamic datasets, Easi3R is plug-and-play. On DAVIS-17 segmentation without SAM2, Easi3Rmonst3r scores 56.5 JM versus DAS3R's 43.5.
What are the practical uses of these findings?
The method benefits novel view synthesis, AR/VR, autonomous navigation, and robotics by enabling robust 4D reconstruction from casual dynamic videos without retraining. The paper demonstrates benchmark results but documents no production deployment.
What limitations do the authors acknowledge?
Easi3R can fail when backbones produce inaccurate depth. While it improves global completeness and distance metrics, a gap remains in per-view depth accuracy; the authors leave per-view depth correction for future work.