arXiv:2511.16361v1 Announce Type: new 
Abstract: Recent guided depth super-resolution methods are premised on the assumption of strictly spatial alignment between depth and RGB, achieving high-quality depth reconstruction. However, in real-world scenarios, the acquisition of strictly aligned RGB-D is hindered by inherent hardware limitations (e.g., physically separate RGB-D sensors) and unavoidable calibration drift induced by mechanical vibrations or temperature variations. Consequently, existing approaches often suffer inevitable performance degradation when applied to misaligned real-world scenes. In this paper, we propose the Multi-Order Matching Network (MOMNet), a novel alignment-free framework that adaptively retrieves and selects the most relevant information from misaligned RGB. Specifically, our method begins with a multi-order matching mechanism, which jointly performs zero-order, first-order, and second-order matching to comprehensively identify RGB information consistent with depth across multi-order feature spaces. To effectively integrate the retrieved RGB and depth, we further introduce a multi-order aggregation composed of multiple structure detectors. This strategy uses multi-order priors as prompts to facilitate the selective feature transfer from RGB to depth. Extensive experiments demonstrate that MOMNet achieves state-of-the-art performance and exhibits outstanding robustness.

تقدم شبكة المطابقة متعددة الأوامر (MOMNet) إطارًا جديدًا لتحسين دقة العمق بدون الحاجة إلى محاذاة مكانية صارمة بين بيانات RGB وعمق. تتناول هذه الطريقة التحديات التي تطرحها قيود الأجهزة والانحراف في المعايرة في السيناريوهات الواقعية، والتي غالبًا ما تؤدي إلى تدهور أداء الأساليب الحالية. من خلال استخدام آلية مطابقة متعددة الأوامر، تسترجع MOMNet بشكل فعال المعلومات ذات الصلة من بيانات RGB غير المتوافقة لتحسين جودة إعادة بناء العمق.

La Red de Coincidencia de Múltiples Órdenes (MOMNet) presenta un marco novedoso para la super-resolución de profundidad que opera sin necesidad de un alineamiento espacial estricto entre los datos RGB y de profundidad. Este enfoque aborda los desafíos planteados por las limitaciones de hardware y la deriva de calibración en escenarios del mundo real, que a menudo degradan el rendimiento de los métodos existentes. Al emplear un mecanismo de coincidencia de múltiples órdenes, MOMNet recupera eficazmente información relevante de datos RGB desalineados para mejorar la calidad de la reconstrucción …

Le Multi-Order Matching Network (MOMNet) propose un cadre novateur pour la super-résolution de profondeur qui fonctionne sans nécessiter un alignement spatial strict entre les données RGB et de profondeur. Cette approche répond aux défis posés par les limitations matérielles et le dérive de calibration dans des scénarios réels, qui dégradent souvent les performances des méthodes existantes. En utilisant un mécanisme de correspondance multi-ordre, MOMNet récupère efficacement des informations pertinentes à partir de données RGB mal alignées pour améliorer la qualité de la reconstruction de prof…

The Multi-Order Matching Network (MOMNet) presents a novel framework for depth super-resolution that operates without the need for strict spatial alignment between RGB and depth data. This approach addresses the challenges posed by hardware limitations and calibration drift in real-world scenarios, which often degrade the performance of existing methods. By employing a multi-order matching mechanism, MOMNet effectively retrieves relevant information from misaligned RGB data to enhance depth reconstruction quality.

Multi-Order Matching Network for Alignment-Free Depth Super-Resolution

arXiv:2511.17092v1 Announce Type: new 
Abstract: Articulated objects are ubiquitous in daily environments, and their 3D reconstruction holds great significance across various fields. However, existing articulated object reconstruction methods typically require costly inputs such as multi-stage and multi-view observations. To address the limitations, we propose a category-agnostic articulated object reconstruction framework via planar Gaussian Splatting, which only uses sparse-view RGB images from a single state. Specifically, we first introduce a Gaussian information field to perceive the optimal sparse viewpoints from candidate camera poses. Then we compress 3D Gaussians into planar Gaussians to facilitate accurate estimation of normal and depth. The planar Gaussians are optimized in a coarse-to-fine manner through depth smooth regularization and few-shot diffusion. Moreover, we introduce a part segmentation probability for each Gaussian primitive and update them by back-projecting part segmentation masks of renderings. Extensive experimental results demonstrate that our method achieves higher-fidelity part-level surface reconstruction on both synthetic and real-world data than existing methods. Codes will be made publicly available.

تم تقديم إطار جديد لإعادة بناء الكائنات المفصلية، باستخدام تقنية التشتت الغاوسي المسطح لإنشاء نماذج ثلاثية الأبعاد من صور RGB ذات عرض ضئيل تم التقاطها من حالة واحدة. تتجاوز هذه الطريقة المبتكرة القيود المفروضة على الأساليب التقليدية التي تتطلب ملاحظات متعددة المراحل ووجهات نظر متعددة، مما يجعل إعادة البناء ثلاثي الأبعاد أكثر سهولة وكفاءة.

Se ha introducido un nuevo marco para la reconstrucción de objetos articulados, utilizando el splatting gaussiano plano para crear modelos 3D a partir de imágenes RGB de vista escasa capturadas desde un solo estado. Este enfoque innovador supera las limitaciones de los métodos tradicionales que requieren observaciones de múltiples vistas, haciendo que la reconstrucción 3D sea más accesible y eficiente.

Un nouveau cadre pour la reconstruction d'objets articulés a été introduit, utilisant le splatting gaussien plan pour créer des modèles 3D à partir d'images RGB à vue sparse capturées à partir d'un seul état. Cette approche innovante surmonte les limitations des méthodes traditionnelles qui nécessitent des observations multi-vues étendues, rendant la reconstruction 3D plus accessible et efficace.

A new framework for articulated object reconstruction has been introduced, utilizing planar Gaussian Splatting to create 3D models from sparse-view RGB images captured from a single state. This innovative approach overcomes the limitations of traditional methods that require extensive multi-view observations, making 3D reconstruction more accessible and efficient.

SPAGS: Sparse-View Articulated Object Reconstruction from Single State via Planar Gaussian Splatting

arXiv:2511.16567v2 Announce Type: replace 
Abstract: In this paper, we introduce POMA-3D, the first self-supervised 3D representation model learned from point maps. Point maps encode explicit 3D coordinates on a structured 2D grid, preserving global 3D geometry while remaining compatible with the input format of 2D foundation models. To transfer rich 2D priors into POMA-3D, a view-to-scene alignment strategy is designed. Moreover, as point maps are view-dependent with respect to a canonical space, we introduce POMA-JEPA, a joint embedding-predictive architecture that enforces geometrically consistent point map features across multiple views. Additionally, we introduce ScenePoint, a point map dataset constructed from 6.5K room-level RGB-D scenes and 1M 2D image scenes to facilitate large-scale POMA-3D pretraining. Experiments show that POMA-3D serves as a strong backbone for both specialist and generalist 3D understanding. It benefits diverse tasks, including 3D question answering, embodied navigation, scene retrieval, and embodied localization, all achieved using only geometric inputs (i.e., 3D coordinates). Overall, our POMA-3D explores a point map way to 3D scene understanding, addressing the scarcity of pretrained priors and limited data in 3D representation learning. Project Page: https://matchlab-imperial.github.io/poma3d/

تم تقديم POMA-3D كنموذج تمثيل ثلاثي الأبعاد ذاتي الإشراف الأول الذي يتعلم من خرائط النقاط، التي ترمز إلى إحداثيات ثلاثية الأبعاد صريحة على شبكة ثنائية الأبعاد منظمة. يحافظ هذا النموذج على الهندسة الثلاثية الأبعاد العالمية ويتوافق مع نماذج الأساس ثنائية الأبعاد، مستخدمًا استراتيجية محاذاة العرض إلى المشهد لتعزيز قدراته.

Se ha presentado POMA-3D como el primer modelo de representación 3D auto-supervisado que aprende a partir de mapas de puntos, que codifican coordenadas 3D explícitas en una cuadrícula 2D estructurada. Este modelo preserva la geometría 3D global y es compatible con modelos de base 2D, utilizando una estrategia de alineación vista-escena para mejorar sus capacidades.

POMA-3D a été introduit comme le premier modèle de représentation 3D auto-supervisé apprenant à partir de cartes de points, qui codent des coordonnées 3D explicites sur une grille 2D structurée. Ce modèle préserve la géométrie 3D globale et est compatible avec les modèles de base 2D, utilisant une stratégie d'alignement vue-scène pour améliorer ses capacités.

POMA-3D has been introduced as the first self-supervised 3D representation model that learns from point maps, which encode explicit 3D coordinates on a structured 2D grid. This model preserves global 3D geometry and is compatible with 2D foundation models, utilizing a view-to-scene alignment strategy to enhance its capabilities.

Multi-Order Matching Network for Alignment-Free Depth Super-Resolution

Was this article worth reading? Share it

Lenso.ai

4o Image Gen

Deptho.ai