homography estimation

This study addresses the challenge of achieving high-precision camera guidance and alignment for multiple rectangular planar regions under extremely limited annotation—requiring only a single labeled image. To this end, the authors propose a geometry-centric intra-image navigation framework that leverages homography as the central organizing variable to unify modeling, alignment, and evaluation. The method integrates intra-image augmentation to generate synthetic training data and employs a two-stage inference mechanism—comprising global detection followed by local refinement—alongside a Stable Warp training strategy. This approach substantially improves alignment accuracy even with low-resolution inputs and enables sparse keypoint localization together with sample-level confidence estimation. The work establishes a robust foundation for geometry-driven camera guidance and self-supervised learning in unconstrained video settings.

This work proposes a lightweight, texture- and pose-prior-free 6-DoF camera relocalization method tailored for structured indoor environments. It introduces 3D planar primitives as region-level structural semantic representations and establishes cross-modal correspondences between query images and a sparse plane-based map through a deep matcher operating in a unified embedding space. Camera poses are then recovered via robust optimization, eliminating the need for photorealistic textures, initial pose estimates, or scene-specific training. Evaluated on multiple benchmarks including ScanNet and 12Scenes, the framework achieves high accuracy and efficiency, demonstrating the effectiveness and generalizability of a structure-primitive-based relocalization paradigm.

This work addresses the challenge of long-term failure in planar object tracking caused by drastic appearance changes or occlusions. We propose a novel approach that integrates SAM 2 segmentation with 8-degree-of-freedom homography estimation. For the first time, SAM 2 is introduced into re-detection for planar tracking, and we innovatively estimate homography directly from the contour of segmentation masks, significantly enhancing robustness. By incorporating this re-detection mechanism into an improved WOFT framework, our method achieves state-of-the-art performance, outperforming the second-best method by 12.4 and 15.2 percentage points in terms of p@15 on the POT-210 and PlanarTrack benchmarks, respectively. Additionally, we provide more accurate initial pose annotations for the PlanarTrack dataset.

Cross-domain unconstrained image matching is highly challenging due to severe appearance, scale, and viewpoint variations; conventional 2D matching methods lack robustness, while existing 3D foundation models, though spatially consistent, exhibit limited capacity for fine-grained geometric modeling—especially under large viewpoint changes. To address this, we propose the first general-purpose matching framework that jointly leverages 2D local semantics and 3D epipolar geometry. Our architecture features a shared backbone with dual-branch matching heads: one for pixel-level correspondence estimation and another for depth-aware spatial constraint modeling. We further introduce linear probing to systematically evaluate the cross-domain generalization of multi-modal vision foundation models and construct the first benchmark specifically designed for extreme viewpoint variation in cross-domain matching. Experiments demonstrate that our method significantly outperforms state-of-the-art approaches on this new benchmark, achieving superior robustness and strong generalization across diverse scenes.

This work addresses the failure of traditional Structure-from-Motion (SfM) methods in planar scenes due to degeneracy in epipolar geometry. The authors propose a unified framework that leverages homography graph embedding to actively exploit planar structures as geometric constraints. By decomposing multi-view homographies induced by dominant planes, relative camera poses are estimated and organized into a pose graph. Global optimization is then achieved through spectral embedding combined with a maximum spanning tree for consensus maximization. The method significantly outperforms existing SfM approaches in strongly planar environments—such as basketball courts—and achieves state-of-the-art or comparable performance on the IMC Phototourism outdoor benchmark, demonstrating robust camera pose recovery in both degenerate and non-degenerate scenarios.

This work addresses the lack of explicit modeling of co-visible regions in image correspondence estimation under large viewpoint and scale variations by proposing a structured feature matching method grounded in co-visibility modeling. It extends the Segment Anything Model (SAM) to multi-view correspondence inference for the first time, leveraging predicted cross-view co-visible masks and bounding boxes as structured priors. A symmetric cross-view interaction mechanism is introduced to enable bidirectional feature exchange and semantic alignment. By integrating mask–box consistency constraints with a unified supervision strategy, the approach shifts the matching paradigm from pixel-level to region-level. The method achieves significant performance gains over existing techniques across multiple challenging benchmarks, demonstrating notably enhanced robustness under extreme viewpoint and scale changes.

This study addresses the problem of recovering true correspondences from multiple noisy and independently permuted point clouds in high-dimensional space. While single-view observation exhibits an information-theoretic impossibility threshold—where exact matching becomes infeasible when the signal strength parameter $b < 2$—this work demonstrates for the first time that incorporating multiple views circumvents this fundamental limitation. Leveraging a high-dimensional Gaussian model and tools from random matrix theory, the authors devise a polynomial-time algorithm that, given $K$ views, achieves near-perfect recovery with only $o(n)$ mismatches whenever $b > K/(K-1)$. Notably, with three views, the method enables efficient and accurate matching in the regime $3/2 < b < 2$, which is provably impossible under a single view.

This work addresses the failure of conventional SIFT-based image registration in scenes dominated by strong linear structures, where local features become ambiguous and poorly discriminative. To overcome this limitation, the authors propose a novel approach that, for the first time, transfers SIFT descriptors into Hough space for matching. By leveraging the Hough transform, linear structures are mapped to prominent peaks, thereby restoring the distinctiveness of features. The resulting Hough-space feature matching framework significantly outperforms standard SIFT in highly structured environments while maintaining comparable registration accuracy in general scenes. This method effectively mitigates the performance degradation of SIFT in structured settings, offering a robust solution for reliable image alignment across diverse scenarios.

This work addresses the challenge of accurately registering RGB-only egocentric video sequences of the same indoor environment captured at different times, under conditions of motion blur, rapid camera movement, and partial scene overlap. The authors propose a zero-shot method that requires no training, depth sensors, or manual annotations, leveraging pretrained vision-language models (VLMs) and off-the-shelf foundation models to convert RGB sequences into object-level 3D scene graphs. Cross-sequence object instances are matched via semantic prompting, augmented with object height priors and intra- and inter-query verification, followed by geometric consistency optimization to refine the rigid transformation. This study presents the first application of VLMs to egocentric scene registration, outperforming both traditional geometric and learning-based approaches on the Aria Digital Twin and Aria Everyday Activities benchmarks, while producing scene graphs directly transferable to downstream tasks.

This work addresses the challenging problem of 6D object pose estimation for unseen objects in open-world scenarios, where severe occlusions, large viewpoint variations, and sparse or absent pose annotations in reference images are prevalent. To tackle this, the authors propose PANY, a unified model-free framework that operates with only one or a few pose-annotated reference images. PANY leverages a multi-view Transformer-based geometric backbone to learn view-consistent structures and cross-view alignment cues, and incorporates pose-graph regularized registration to fuse geometric information when auxiliary views are available. Breaking away from conventional pairwise matching paradigms, PANY supports arbitrary reference settings with either RGB or RGB-D inputs. Experiments demonstrate that PANY significantly outperforms existing model-free methods on the YCB-Video and LineMOD-Occlusion benchmarks, achieving pose accuracy improvements of 12% and over 20%, respectively, while exhibiting strong robustness in real-world scenes.