Online3R: Online Learning for Consistent Sequential Reconstruction Based on Geometry Foundation Model

Shunkai Zhou; Zike Yan; Fei Xue; Dong Wu; Yuchen Deng; Hongbin Zha

cs.CVApr 10, 2026

Online3R: Online Learning for Consistent Sequential Reconstruction Based on Geometry Foundation Model

Authors: Shunkai Zhou, Zike Yan, Fei Xue, Dong Wu, Yuchen Deng, Hongbin Zha

Organizations: School of Intelligence Science and Technology, Peking University · State Key Laboratory of General Artificial Intelligence · T-Stone Robotics Institute, The Chinese University of Hong Kong · NVIDIA ·

Abstract

We present Online3R, a new sequential reconstruction framework that is capable of adapting to new scenes through online learning, effectively resolving inconsistency issues. Specifically, we introduce a set of learnable lightweight visual prompts into a pretrained, frozen geometry foundation model to capture the knowledge of new environments while preserving the fundamental capability of the foundation model for geometry prediction. To solve the problems of missing groundtruth and the requirement of high efficiency when updating these visual prompts at test time, we introduce a local-global self-supervised learning strategy by enforcing the local and global consistency constraints on predictions. The local consistency constraints are conducted on intermediate and previously local fused results, enabling the model to be trained with high-quality pseudo groundtruth signals; the global consistency constraints are operated on sparse keyframes spanning long distances rather than per frame, allowing the model to learn from a consistent prediction over a long trajectory in an efficient way. Our experiments demonstrate that Online3R outperforms previous state-of-the-art methods on various benchmarks. Project page: https://shunkaizhou.github.io/online3r-1.0/

Figures & tables

Figure 1: Overview of our proposed Online3R. The core of our Online3R lies in constructing self-supervised methods and online prompt tuning, enabling the model to adapt to the current scene and ensuring consistent reconstruction results. We leverage a local consistency loss derived from temporally fused geometry to enhance the accuracy of subsequent predictions, and a global consistency loss that enforces geometric invariance across distant views to mitigate long-term drift, guaranteeing both consistency and efficiency.

Type	Methods	360	desk	desk2	floor	plant	room	rpy	teddy	xyz	avg
Calibrated	ORB-SLAM3 [ 2 ]	–	0.017	0.210	–	0.034	–	–	–	0.009	–
	DeepV2D [ 24 ]	0.243	0.166	0.379	1.653	0.203	0.246	0.105	0.316	0.064	0.375
	DeepFactors [ 3 ]	0.159	0.170	0.253	0.169	0.305	0.364	0.043	0.601	0.035	0.233
	DPV-SLAM [ 12 ]	0.112	0.018	0.029	0.057	0.021	0.330	0.030	0.084	0.010	0.076
	DPV-SLAM++ [ 12 ]	0.132	0.018	0.029	0.050	0.022	0.096	0.032	0.098	0.010	0.054
	GO-SLAM [ 36 ]	0.089	0.016	0.028	0.025	0.026	0.052	0.019	0.048	0.010	0.035

Table 1: Absolute trajectory error (ATE (m)

\downarrow

) on TUM RGB-D [ 23 ] . We evaluate methods with and without ground-truth camera intrinsics as inputs (noted as “Calibrated” and “Uncalibrated”). The best and the second best results of each type are highlighted.

Methods	brea	comp	gree	grey	kitc	morn	stai	thin	whi	avg
Spann3R	1.464	1.980	0.981	1.269	1.989	0.631	1.997	0.368	2.319	1.444
CUT3R	0.566	1.450	0.868	0.922	1.099	0.329	0.466	0.204	1.847	0.861
Point3R	0.416	1.155	0.302	0.659	1.354	0.269	0.657	0.133	0.593	0.615
M-S*	0.109	0.079	0.095	0.089	0.176	0.058	0.056	0.046	0.105	0.090
Ours*	0.102	0.063	0.084	0.102	0.118	0.057	0.047	0.047	0.066	0.076

Table 2: Absolute trajectory error (ATE (m)

\downarrow

) on NRGBD [ 1 ] . MASt3R-SLAM* [ 16 ] is abbreviated as M-S*. The best and the second best results are highlighted.

Types	Methods	7-scenes			NRGBD
Types	Methods	Acc $\downarrow$	Comp $\downarrow$	Chamf $\downarrow$	Acc $\downarrow$	Comp $\downarrow$	Chamf $\downarrow$
Offline	DUSt3R-GA [ 30 ]	0.146	0.181	0.164	0.144	0.154	0.149
	MASt3R-GA [ 10 ]	0.185	0.180	0.183	0.085	0.063	0.074
	MonST3R-GA [ 35 ]	0.248	0.266	0.257	0.272	0.287	0.280
	Test3R [ 34 ]	0.105	0.136	0.121	0.083	0.079	0.081
Online	Spann3R [ 27 ]	0.298	0.205	0.252	0.416	0.417	0.417

Table 3: Quantitative 3D reconstruction results on 7-scenes and NRGBD datasets. We categorize methods into “Offline” and “Online” [ 29 ] , and use “GA” to mark methods with global alignment. The best and the second best results of all methods are highlighted. Our method achieves competitive or better performance than those offline or online methods.

Figure 2: Qualitative Comparison on 3D Reconstruction Consistency. We present reconstruction results on two sequences, with 7-Scenes-heads on the left and NRGBD-staircase on the right, separated by dashed lines. The first row shows the global point cloud reconstruction from a far viewpoint. The second row zooms in to view the near viewpoint. The third row highlights details using bounding boxes. It is evident that our method achieves better consistency compared to MASt3R-SLAM*.

Figure 3: Qualitative results for reconstruction of non-overlapping image pairs.

Table 7

Streaming 3D Reconstruction

Streaming 3D reconstruction incrementally builds geometrically consistent scene maps from incoming images or video under tight memory and latency limits.

See all 99 papers

Free Geometry: Refining 3D Reconstruction from Longer Versions of Itself
Apr 15, 2026Yuhang Dai, Xingyi YangFeed-Forward 3D Reconstruction Multi-View 3D Reconstruction
The Hong Kong Polytechnic University, Hong Kong SAR
Scal3R: Learning Efficient Multi-Relative Pose Query for Scalable Online 3D Reconstruction
Sep 3, 2026Chin-Yang Lin, Yang-Che Sun, Cheng Sun +5Relative Pose Estimation Streaming 3D Reconstruction
Department of Computer Science, National Yang Ming Chiao Tung University · NVIDIA
Glob3R: Global Structure-from-Motion with 3D Foundation Models
Jul 10, 2026Junyuan Deng, Heng Li, Kejie Qiu +7Bundle Adjustment Multi-View 3D Reconstruction
The Hong Kong University of Science and Technology · Tongyi Lab, Alibaba Group · Nanjing University +1

Apr 15, 2026J/K move · Enter open · S save

Free Geometry: Refining 3D Reconstruction from Longer Versions of Itself

Yuhang Dai, Xingyi Yang

The Hong Kong Polytechnic University, Hong Kong SAR

Feed-Forward 3D Reconstruction Multi-View 3D Reconstruction Test-Time Adaptation Self-Supervised Learning

Open paper AskPDF arXiv

Feed-forward 3D reconstruction models are efficient but rigid: once trained, they perform inference in a zero-shot manner and cannot adapt to the test scene. As a result, visually plausible reconstructions often contain errors, particularly under occlusions, specularities, and ambiguous cues. To address this, we introduce Free Geometry, a framework that enables feed-forward 3D reconstruction models to self-evolve at test time without any 3D ground truth. Our key insight is that, when the model receives more views, it produces more reliable and view-consistent reconstructions. Leveraging this property, given a testing sequence, we mask a subset of frames to construct a self-supervised task. Free Geometry enforces cross-view feature consistency between representations from full and partial observations, while maintaining the pairwise relations implied by the held-out frames. This self-supervision allows for fast recalibration via lightweight LoRA updates, taking less than 2 minutes per dataset on a single GPU. Our approach consistently improves state-of-the-art foundation models, including Depth Anything 3 and VGGT, across 4 benchmark datasets, yielding an average improvement of 3.73% in camera pose accuracy and 2.88% in point map prediction. Code is available at https://github.com/hiteacherIamhumble/Free-Geometry .

Abstract

Streaming 3D Reconstruction

Explore similar work

Free Geometry: Refining 3D Reconstruction from Longer Versions of Itself

Scal3R: Learning Efficient Multi-Relative Pose Query for Scalable Online 3D Reconstruction

Glob3R: Global Structure-from-Motion with 3D Foundation Models