arXiv:2510.21432v1 Announce Type: new 
Abstract: We propose ArtiLatent, a generative framework that synthesizes human-made 3D objects with fine-grained geometry, accurate articulation, and realistic appearance. Our approach jointly models part geometry and articulation dynamics by embedding sparse voxel representations and associated articulation properties, including joint type, axis, origin, range, and part category, into a unified latent space via a variational autoencoder. A latent diffusion model is then trained over this space to enable diverse yet physically plausible sampling. To reconstruct photorealistic 3D shapes, we introduce an articulation-aware Gaussian decoder that accounts for articulation-dependent visibility changes (e.g., revealing the interior of a drawer when opened). By conditioning appearance decoding on articulation state, our method assigns plausible texture features to regions that are typically occluded in static poses, significantly improving visual realism across articulation configurations. Extensive experiments on furniture-like objects from PartNet-Mobility and ACD datasets demonstrate that ArtiLatent outperforms existing approaches in geometric consistency and appearance fidelity. Our framework provides a scalable solution for articulated 3D object synthesis and manipulation.

ArtiLatent هو إطار توليدي مبتكر ينشئ كائنات ثلاثية الأبعاد واقعية مع هندسة مفصلة وحركة مفصلية. من خلال دمج هندسة الأجزاء وديناميات الحركة في فضاء كامن موحد، فإنه يعزز من واقعية ووظائف النماذج ثلاثية الأبعاد. هذه الخطوة مهمة للصناعات مثل الألعاب والرسوم المتحركة والواقع الافتراضي، حيث تعتبر التمثيلات ثلاثية الأبعاد عالية الجودة أمرًا حيويًا.

ArtiLatent es un marco generativo innovador que crea objetos 3D realistas con geometría y articulación detalladas. Al combinar la geometría de las partes y la dinámica de articulación en un espacio latente unificado, mejora el realismo y la funcionalidad de los modelos 3D. Este avance es significativo para industrias como los videojuegos, la animación y la realidad virtual, donde las representaciones 3D de alta calidad son cruciales.

ArtiLatent est un cadre génératif innovant qui crée des objets 3D réalistes avec une géométrie et une articulation détaillées. En combinant la géométrie des parties et la dynamique d'articulation dans un espace latent unifié, il améliore le réalisme et la fonctionnalité des modèles 3D. Cette avancée est significative pour des secteurs comme le jeu vidéo, l'animation et la réalité virtuelle, où des représentations 3D de haute qualité sont cruciales.

ArtiLatent is an innovative generative framework that creates realistic 3D objects with detailed geometry and articulation. By combining part geometry and articulation dynamics into a unified latent space, it enhances the realism and functionality of 3D models. This advancement is significant for industries like gaming, animation, and virtual reality, where high-quality 3D representations are crucial.

ArtiLatent: Realistic Articulated 3D Object Generation via Structured Latents

arXiv:2512.04786v1 Announce Type: new 
Abstract: Generating high-fidelity, seamless textures directly on 3D surfaces, what we term 3D-native texturing, remains a fundamental open challenge, with the potential to overcome long-standing limitations of UV-based and multi-view projection methods. However, existing native approaches are constrained by the absence of a powerful and versatile latent representation, which severely limits the fidelity and generality of their generated textures. We identify this representation gap as the principal barrier to further progress. We introduce LaFiTe, a framework that addresses this challenge by learning to generate textures as a 3D generative sparse latent color field. At its core, LaFiTe employs a variational autoencoder (VAE) to encode complex surface appearance into a sparse, structured latent space, which is subsequently decoded into a continuous color field. This representation achieves unprecedented fidelity, exceeding state-of-the-art methods by >10 dB PSNR in reconstruction, by effectively disentangling texture appearance from mesh topology and UV parameterization. Building upon this strong representation, a conditional rectified-flow model synthesizes high-quality, coherent textures across diverse styles and geometries. Extensive experiments demonstrate that LaFiTe not only sets a new benchmark for 3D-native texturing but also enables flexible downstream applications such as material synthesis and texture super-resolution, paving the way for the next generation of 3D content creation workflows.

تم تقديم LaFiTe كإطار جديد لتوليد القوام عالي الدقة والسلس مباشرة على الأسطح ثلاثية الأبعاد، مما يعالج القيود المفروضة على الطرق التقليدية المعتمدة على UV وعرض المناظر المتعددة. من خلال استخدام مشفر متغير (VAE)، يقوم LaFiTe بتشفير المظاهر المعقدة للأسطح في فضاء كامن منظم، والذي يتم فك تشفيره بعد ذلك إلى حقل لون مستمر، مما يحقق دقة غير مسبوقة في القوام.

LaFiTe se ha presentado como un nuevo marco para generar texturas de alta fidelidad y sin costuras directamente en superficies 3D, abordando las limitaciones de los métodos tradicionales basados en UV y proyección de múltiples vistas. Al utilizar un autoencoder variacional (VAE), LaFiTe codifica apariencias complejas de superficie en un espacio latente estructurado, que luego se decodifica en un campo de color continuo, logrando una fidelidad de textura sin precedentes.

LaFiTe a été introduit comme un nouveau cadre pour générer des textures de haute fidélité et sans couture directement sur des surfaces 3D, répondant aux limitations des méthodes traditionnelles basées sur les UV et la projection multi-vues. En utilisant un autoencodeur variationnel (VAE), LaFiTe encode des apparences de surface complexes dans un espace latent structuré, qui est ensuite décodé en un champ de couleur continu, atteignant une fidélité de texture sans précédent.

LaFiTe has been introduced as a novel framework for generating high-fidelity, seamless textures directly on 3D surfaces, addressing the limitations of traditional UV-based and multi-view projection methods. By utilizing a variational autoencoder (VAE), LaFiTe encodes complex surface appearances into a structured latent space, which is then decoded into a continuous color field, achieving unprecedented texture fidelity.

LaFiTe: A Generative Latent Field for 3D Native Texturing

arXiv:2512.03525v1 Announce Type: new 
Abstract: Compressed sensing enables sparse sampling but relies on generic bases and random measurements, limiting efficiency and reconstruction quality. Optimal sensor placement uses historcal data to design tailored sampling patterns, yet its fixed, linear bases cannot adapt to nonlinear or sample-specific variations. Generative model-based compressed sensing improves reconstruction using deep generative priors but still employs suboptimal random sampling. We propose an adaptive sparse sensing framework that couples a variational autoencoder prior with reinforcement learning to select measurements sequentially. Experiments show that this approach outperforms CS, OSP, and Generative model-based reconstruction from sparse measurements.

تم اقتراح إطار جديد لأخذ العينات التكيفية، يجمع بين مشفر متغير التعلم مع التعلم المعزز لتحسين اختيار القياسات في الاستشعار المضغوط. تعالج هذه الطريقة قيود تقنيات أخذ العينات التقليدية من خلال السماح باختيار القياسات بشكل متسلسل استنادًا إلى البيانات التاريخية، مما يؤدي إلى تحسين جودة إعادة البناء والكفاءة.

Se ha propuesto un nuevo marco de muestreo adaptativo que integra un autoencoder variacional con aprendizaje por refuerzo para mejorar la selección de medidas en el muestreo comprimido. Este método aborda las limitaciones de las técnicas de muestreo tradicionales al permitir elecciones de medidas secuenciales basadas en datos históricos, lo que conduce a una mejor calidad de reconstrucción y eficiencia.

Un nouveau cadre d'échantillonnage adaptatif a été proposé, intégrant un autoencodeur variationnel avec un apprentissage par renforcement pour améliorer la sélection des mesures dans le cadre du sensing compressé. Cette méthode répond aux limitations des techniques d'échantillonnage traditionnelles en permettant des choix de mesures séquentiels basés sur des données historiques, ce qui conduit à une meilleure qualité de reconstruction et à une efficacité accrue.

A new adaptive sampling framework has been proposed, integrating a variational autoencoder with reinforcement learning to enhance measurement selection in compressed sensing. This method addresses limitations of traditional sampling techniques by allowing for sequential measurement choices based on historical data, leading to improved reconstruction quality and efficiency.

Adaptive sampling using variational autoencoder and reinforcement learning

arXiv:2512.02032v1 Announce Type: cross 
Abstract: Structural connectomes are detailed graphs that map how different brain regions are physically connected, offering critical insight into aging, cognition, and neurodegenerative diseases. However, these connectomes are high-dimensional and densely interconnected, which makes them difficult to interpret and analyze at scale. While low-dimensional spaces like PCA and autoencoders are often used to capture major sources of variation, their latent spaces are generally continuous and cannot fully reflect the mixed nature of variability in connectomes, which include both continuous (e.g., connectivity strength) and discrete factors (e.g., imaging site). Motivated by this, we propose a variational autoencoder (VAE) with a hybrid latent space that jointly models the discrete and continuous components. We analyze a large dataset of 5,761 connectomes from six Alzheimer's disease studies with ten acquisition protocols. Each connectome represents a single scan from a unique subject (3579 females, 2182 males), aged 22 to 102, with 4338 cognitively normal, 809 with mild cognitive impairment (MCI), and 614 with Alzheimer's disease (AD). Each connectome contains 121 brain regions defined by the BrainCOLOR atlas. We train our hybrid VAE in an unsupervised way and characterize what each latent component captures. We find that the discrete space is particularly effective at capturing subtle site-related differences, achieving an Adjusted Rand Index (ARI) of 0.65 with site labels, significantly outperforming PCA and a standard VAE followed by clustering (p < 0.05). These results demonstrate that the hybrid latent space can disentangle distinct sources of variability in connectomes in an unsupervised manner, offering potential for large-scale connectome analysis.

تم اقتراح نهج جديد يستخدم مشفرًا متغيرًا (VAE) مع مساحة كامنة هجينة لتحليل الشبكات الهيكلية، وهي رسوم بيانية معقدة تمثل الروابط بين مناطق الدماغ. تتناول هذه الطريقة التحديات المتعلقة بتفسير البيانات عالية الأبعاد، لا سيما في سياق مرض الزهايمر، من خلال نمذجة العوامل المستمرة والقطعية في الشبكات الهيكلية بشكل فعال.

Se ha propuesto un enfoque novedoso que utiliza un autoencoder variacional (VAE) con un espacio latente híbrido para analizar conectomas estructurales, que son gráficos complejos que representan las conexiones entre las regiones del cerebro. Este método aborda los desafíos de interpretar datos de alta dimensión, especialmente en el contexto de la enfermedad de Alzheimer, al modelar de manera efectiva tanto los factores continuos como discretos en los conectomas.

Une nouvelle approche utilisant un autoencodeur variationnel (VAE) avec un espace latent hybride a été proposée pour analyser les connectomes structurels, qui sont des graphes complexes représentant les connexions entre les régions du cerveau. Cette méthode répond aux défis d'interprétation des données de haute dimension, en particulier dans le contexte de la maladie d'Alzheimer, en modélisant efficacement à la fois les facteurs continus et discrets dans les connectomes.

A novel approach utilizing a variational autoencoder (VAE) with a hybrid latent space has been proposed to analyze structural connectomes, which are complex graphs mapping brain region connections. This method addresses the challenges of interpreting high-dimensional data, particularly in the context of Alzheimer's disease, by effectively modeling both continuous and discrete factors in connectomes.

ArtiLatent: Realistic Articulated 3D Object Generation via Structured Latents

Was this article worth reading? Share it

LucidQuery AI

Artefacts.ai

Lovart-Al Ip Avatar Design