gravity direction estimation

This work addresses the limitations of traditional 3D reconstruction methods, which predict point maps in camera-centered coordinates, struggle to incorporate scene structural priors, and suffer from high rotational degrees of freedom across views, leading to inconsistent reconstructions. To overcome these issues, the authors propose predicting point maps in a gravity-aligned upright coordinate system, thereby reducing inter-view rotational ambiguity through a shared vertical axis. They introduce the Gravity Grounded Geometry Transformer (G3T) model and the G3T-Long incremental reconstruction framework, which for the first time integrate gravity-aligned coordinates into point map prediction by combining a Transformer architecture, gravity-aware pose estimation, and a submap stitching strategy. Experiments demonstrate that this approach significantly improves reconstruction accuracy and robustness, outperforming existing methods in incremental 3D reconstruction and validating the effectiveness of gravity-aligned representations.

This work addresses the limitation of existing RGB-based multi-view reconstruction methods, which produce monolithic scene representations lacking explicit physical structure and thus hinder stable physical interaction. The authors propose an end-to-end reconstruction framework that relies solely on RGB images and introduces gravity as a universal physical prior. By aligning views in a gravity-consistent coordinate system, reconstructing object-centric rigid-body meshes, and employing conditional 3D point classification to remove background redundancy, the method decouples foreground objects from background geometry without requiring CAD model retrieval. The output is a structured hybrid representation suitable for simulation. Experiments demonstrate significant improvements over retrieval-based baselines in 6-DoF object pose accuracy, decoupling quality, and rendering-to-simulation efficiency, both in simulated and real-world scenes.

To address high geometric ambiguity and weak semantic guidance in indoor 3D reconstruction from sparse views, this paper proposes the first end-to-end geometric-semantic co-optimization framework. Methodologically, it integrates semantic priors from 2D foundation models and introduces depth-consistency constraints alongside multi-face normal regularization, enabling semantics to actively guide geometric optimization. By unifying differentiable rendering with multi-view geometry, the framework performs semantic-driven optimization of 3D Gaussian splatting. Evaluated on standard benchmarks including ScanNet, our method achieves state-of-the-art performance in both novel-view synthesis and geometric reconstruction. It significantly improves model completeness under sparse input—reducing Chamfer distance by +12.3%—and enhances geometric fidelity—increasing PSNR by +8.7%. These results empirically validate the substantial benefit of semantic priors in strengthening geometric reconstruction accuracy and robustness.

In large-scale Structure-from-Motion (SfM), sparse inter-view overlap and drastic viewpoint changes—especially in aerial-to-ground scenarios—lead to low cross-image feature matching density and weak geometric consistency. To address this, we propose a geometry-guided hybrid matching paradigm: (1) geometric verification is formulated as an optimization problem based on Sampson distance; (2) detector-agnostic dense matching is fused with detector-driven sparse anchor guidance, where sparse anchors constrain and enhance the geometric consistency of dense matches; and (3) multi-view geometric consistency is explicitly modeled. Our method significantly improves both matching density and accuracy, outperforming state-of-the-art approaches in extreme large-scale settings. Consequently, camera pose estimation becomes more accurate, and the reconstructed 3D point cloud achieves higher completeness and fidelity.

This work addresses the lack of physical plausibility in single-image 3D reconstruction. We propose the first physics-compatible reconstruction framework that enforces static equilibrium as a hard constraint. Methodologically, we explicitly decouple and jointly optimize material stiffness, external loading forces, and the static equilibrium geometry; deformation responses are modeled via differentiable physics simulation, enabling gradient-based joint optimization of all variables. Our approach breaks from conventional simplifications—such as rigid-body assumptions or neglect of external forces—by embedding real-world physical constraints directly into the single-image reconstruction pipeline. Evaluated on Objaverse, our method yields reconstructions with significantly improved mechanical stability, suitable for downstream dynamic simulation and 3D printing. Physical validation via real-world force testing further confirms the structural robustness of the generated models.

Global Structure-from-Motion (SfM) is prone to errors caused by noisy baselines, weak view-graph constraints, and visually ambiguous matches due to its reliance on scale-ambiguous epipolar geometry. This work introduces monocular depth maps into the global SfM pipeline for the first time, proposing a depth-aware reconstruction framework. By leveraging depth priors, epipolar constraints are transformed into scale-aware relative pose constraints. The approach further enhances robustness and accuracy through view-graph filtering, depth-consistent match pruning, and depth-guided initialization of camera poses and 3D points. Evaluated on the ETH3D and IMC2021 benchmarks, the method consistently outperforms existing global SfM techniques under both sparse and dense feature matching frontends.

Existing 3D foundation models suffer from insufficient accuracy when reconstructing large-scale unordered image collections or long sequences, and patch-based processing often leads to drift and inconsistency. This work proposes a novel approach that introduces a lightweight dense matching head on top of a frozen Pi3X backbone to predict deformations between a reference frame and its neighboring views, thereby establishing reliable multi-view feature trajectories. Furthermore, a keyframe sliding window mechanism is designed to integrate the prior knowledge from the 3D foundation model with global geometric optimization, enabling scalable global motion averaging and bundle adjustment. The method significantly outperforms current feed-forward foundation models and scalable reconstruction techniques across indoor, outdoor, large-scale driving, and unordered structure-from-motion benchmarks, achieving notable improvements in reconstruction accuracy, robustness, and neural rendering quality.

This work addresses the challenging alignment problem between generative 3D reconstructions and sparse, noisy monocular observations, which is hindered by scale ambiguity, geometric hallucinations, and initial lack of overlap. The authors propose a training-free geometric alignment framework that recovers metric scale and pose via Sim(3) transformation through a coarse-to-fine strategy for robust initialization and precise refinement. Key innovations include an explicit scale factor to resolve scale ambiguity, a geometry-aware descriptor paired with a decoupled closed-form solver, and a hallucination filtering mechanism to suppress spurious geometry generated by neural models. Evaluated on the newly introduced GenPMOAlign–Where2Place benchmark, the method significantly outperforms both classical geometric and state-of-the-art learning-based approaches, achieving stable and highly accurate alignment.

Existing feedforward 3D foundation models, constrained by central perspective projection, struggle to accommodate the push-broom imaging geometry of satellites, limiting their applicability in multi-view satellite 3D reconstruction. This work proposes a lightweight adaptation framework that requires no fine-tuning of the backbone network: it selects an optimal view sequence through geometric consistency constraints, parameterizes push-broom rays into geometric tokens using rational function models, and introduces a ray-direction-aware adapter to inject these tokens into a frozen Transformer backbone. This approach achieves, for the first time, an effective integration of physical imaging geometry with deep feedforward architectures, significantly enhancing accuracy and robustness in digital surface model (DSM) generation and demonstrating the critical role of explicit geometric embedding and optimized view selection.

This work addresses the lack of metric-scale consistency in monocular foundation models despite their ability to predict dense geometry. The authors propose a visual-inertial guided dense alignment strategy that integrates SVO+IMU visual-inertial odometry with Depth Anything V3 to achieve globally consistent, metric-scale dense monocular reconstruction without requiring ground-truth poses. By leveraging either pose-conditioned or decoupled alignment mechanisms, the method fuses local depth predictions with scale, pose, and world-frame information from an inertial front-end. Experiments on the EuRoC dataset demonstrate that pose injection reduces scale error to approximately 1% and achieves an F@0.10 score of 0.463, while the decoupled hybrid strategy further improves performance to 0.676. Additional evaluation on TUM RGB-D confirms the approach’s generalization capability.