arXiv:2512.06368v2 Announce Type: replace 
Abstract: Monocular dynamic video reconstruction faces significant challenges in dynamic human scenes due to geometric inconsistencies and resolution degradation issues. Existing methods lack 3D human structural understanding, producing geometrically inconsistent results with distorted limb proportions and unnatural human-object fusion, while memory-constrained downsampling causes human boundary drift toward background geometry. To address these limitations, we propose to incorporate hybrid geometric priors that combine SMPL human body models with monocular depth estimation. Our approach leverages structured human priors to maintain surface consistency while capturing fine-grained geometric details in human regions. We introduce HuPrior3R, featuring a hierarchical pipeline with refinement components that processes full-resolution images for overall scene geometry, then applies strategic cropping and cross-attention fusion for human-specific detail enhancement. The method integrates SMPL priors through a Feature Fusion Module to ensure geometrically plausible reconstruction while preserving fine-grained human boundaries. Extensive experiments on TUM Dynamics and GTA-IM datasets demonstrate superior performance in dynamic human reconstruction.

تم تقديم HuPrior3R كنهج جديد لتحسين إعادة البناء الديناميكي ثلاثي الأبعاد من مقاطع الفيديو الأحادية، حيث يتناول تحديات كبيرة مثل التناقضات الهندسية وتدهور الدقة في المشاهد البشرية الديناميكية. تدمج الطريقة أولويات هندسية هجينة، تجمع بين نماذج الجسم البشري SMPL مع تقدير العمق الأحادي لتحسين تناسق السطح والتقاط التفاصيل في المناطق البشرية.

HuPrior3R se ha presentado como un enfoque novedoso para mejorar la reconstrucción dinámica 3D a partir de videos monoculares, abordando desafíos significativos como las inconsistencias geométricas y la degradación de la resolución en escenas humanas dinámicas. El método integra priors geométricos híbridos, combinando modelos corporales humanos SMPL con estimación de profundidad monocular para mejorar la consistencia de la superficie y la captura de detalles en las regiones humanas.

HuPrior3R a été introduit comme une nouvelle approche pour améliorer la reconstruction dynamique 3D à partir de vidéos monoculaires, en s'attaquant à des défis significatifs tels que les incohérences géométriques et la dégradation de la résolution dans des scènes humaines dynamiques. La méthode intègre des priorités géométriques hybrides, combinant des modèles corporels humains SMPL avec une estimation de profondeur monoculaire pour améliorer la cohérence de surface et la capture de détails dans les régions humaines.

HuPrior3R has been introduced as a novel approach to enhance 3D dynamic reconstruction from monocular videos, addressing significant challenges such as geometric inconsistencies and resolution degradation in dynamic human scenes. The method integrates hybrid geometric priors, combining SMPL human body models with monocular depth estimation to improve surface consistency and detail capture in human regions.

HuPrior3R: Incorporating Human Priors for Better 3D Dynamic Reconstruction from Monocular Videos

arXiv:2512.08215v1 Announce Type: new 
Abstract: The creation of lifelike human avatars capable of realistic pose variation and viewpoint flexibility remains a fundamental challenge in computer vision and graphics. Current approaches typically yield either geometrically inconsistent multi-view images or sacrifice photorealism, resulting in blurry outputs under diverse viewing angles and complex motions. To address these issues, we propose Blur2Sharp, a novel framework integrating 3D-aware neural rendering and diffusion models to generate sharp, geometrically consistent novel-view images from only a single reference view. Our method employs a dual-conditioning architecture: initially, a Human NeRF model generates geometrically coherent multi-view renderings for target poses, explicitly encoding 3D structural guidance. Subsequently, a diffusion model conditioned on these renderings refines the generated images, preserving fine-grained details and structural fidelity. We further enhance visual quality through hierarchical feature fusion, incorporating texture, normal, and semantic priors extracted from parametric SMPL models to simultaneously improve global coherence and local detail accuracy. Extensive experiments demonstrate that Blur2Sharp consistently surpasses state-of-the-art techniques in both novel pose and view generation tasks, particularly excelling under challenging scenarios involving loose clothing and occlusions.

يمثل تقديم Blur2Sharp تقدمًا كبيرًا في تركيب الصور الرمزية البشرية الواقعية، حيث يتناول التحديات المتعلقة بإنشاء تنوعات واقعية في الوضع ومرونة الزاوية في رؤية الكمبيوتر والرسوميات. يجمع هذا الإطار الجديد بين تقنية العرض المعتمدة على 3D ونماذج الانتشار لإنتاج صور حادة ومتسقة هندسيًا من عرض مرجعي واحد فقط.

La introducción de Blur2Sharp representa un avance significativo en la síntesis de avatares humanos realistas, abordando los desafíos de generar variaciones de pose y flexibilidad de punto de vista en visión por computadora y gráficos. Este nuevo marco combina renderizado neural consciente de 3D con modelos de difusión para producir imágenes nítidas y geométricamente consistentes a partir de una sola vista de referencia.

L'introduction de Blur2Sharp représente une avancée significative dans la synthèse d'avatars humains réalistes, répondant aux défis de génération de variations de pose et de flexibilité de point de vue en vision par ordinateur et en graphisme. Ce nouveau cadre combine le rendu neural conscient de la 3D avec des modèles de diffusion pour produire des images nettes et géométriquement cohérentes à partir d'une seule vue de référence.

The introduction of Blur2Sharp presents a significant advancement in the synthesis of lifelike human avatars, addressing the longstanding challenges of generating realistic pose variations and viewpoint flexibility in computer vision and graphics. This novel framework combines 3D-aware neural rendering with diffusion models to produce sharp, geometrically consistent images from a single reference view.

Blur2Sharp: Human Novel Pose and View Synthesis with Generative Prior Refinement

arXiv:2512.07459v1 Announce Type: cross 
Abstract: Generating realistic human geometry animations remains a challenging task, as it requires modeling natural clothing dynamics with fine-grained geometric details under limited data. To address these challenges, we propose two novel designs. First, we propose a compact distribution-based latent representation that enables efficient and high-quality geometry generation. We improve upon previous work by establishing a more uniform mapping between SMPL and avatar geometries. Second, we introduce a generative animation model that fully exploits the diversity of limited motion data. We focus on short-term transitions while maintaining long-term consistency through an identity-conditioned design. These two designs formulate our method as a two-stage framework: the first stage learns a latent space, while the second learns to generate animations within this latent space. We conducted experiments on both our latent space and animation model. We demonstrate that our latent space produces high-fidelity human geometry surpassing previous methods ($90\%$ lower Chamfer Dist.). The animation model synthesizes diverse animations with detailed and natural dynamics ($2.2 \times$ higher user study score), achieving the best results across all evaluation metrics.

تم اقتراح نهج جديد لتوليد الرسوم المتحركة للجيومترات البشرية الواقعية، حيث يتناول التحديات المتعلقة بنمذجة ديناميات الملابس والتفاصيل الهندسية مع بيانات محدودة. تقدم هذه الطريقة تمثيلًا كامنًا قائمًا على توزيع مضغوط ونموذج رسوم متحركة توليدي يعزز تنوع بيانات الحركة، وهيكلها في إطار من مرحلتين لتحسين توليد الرسوم المتحركة.

Se ha propuesto un enfoque novedoso para generar animaciones de geometría humana realistas, abordando los desafíos de modelar la dinámica de la ropa y los detalles geométricos con datos limitados. Este método introduce una representación latente basada en una distribución compacta y un modelo de animación generativa que mejora la diversidad de los datos de movimiento, estructurado en un marco de dos etapas para una generación de animaciones mejorada.

Une nouvelle approche pour générer des animations de géométrie humaine réalistes a été proposée, répondant aux défis de la modélisation de la dynamique des vêtements et des détails géométriques avec des données limitées. Cette méthode introduit une représentation latente basée sur une distribution compacte et un modèle d'animation génératif qui améliore la diversité des données de mouvement, structuré dans un cadre en deux étapes pour une génération d'animations améliorée.

A novel approach to generating realistic human geometry animations has been proposed, addressing the challenges of modeling clothing dynamics and geometric details with limited data. This method introduces a compact distribution-based latent representation and a generative animation model that enhances the diversity of motion data, structured in a two-stage framework for improved animation generation.

HuPrior3R: Incorporating Human Priors for Better 3D Dynamic Reconstruction from Monocular Videos

Was this article worth reading? Share it

LucidQuery AI

GPTHumanizer

Deptho.ai