arXiv:2412.19413v2 Announce Type: replace 
Abstract: Consistency Models (CMs) have significantly accelerated the sampling process in diffusion models, yielding impressive results in synthesizing high-resolution images. To explore and extend these advancements to point-cloud-based 3D shape generation, we propose a novel Multi-scale Latent Point Consistency Model (MLPCM). Our MLPCM follows a latent diffusion framework and introduces hierarchical levels of latent representations, ranging from point-level to super-point levels, each corresponding to a different spatial resolution. We design a multi-scale latent integration module along with 3D spatial attention to effectively denoise the point-level latent representations conditioned on those from multiple super-point levels. Additionally, we propose a latent consistency model, learned through consistency distillation, that compresses the prior into a one-step generator. This significantly improves sampling efficiency while preserving the performance of the original teacher model. Extensive experiments on standard benchmarks ShapeNet and ShapeNet-Vol demonstrate that MLPCM achieves a 100x speedup in the generation process, while surpassing state-of-the-art diffusion models in terms of both shape quality and diversity.

تقدم دراسة جديدة نموذج التناسق الكامن متعدد المقاييس (MLPCM) الذي يعزز توليد الأشكال ثلاثية الأبعاد من خلال تطبيق نماذج التناسق المتقدمة المستمدة من توليد الصور. هذه الابتكار مهم لأنه يعد بتحسين كفاءة وجودة توليد الأشكال ثلاثية الأبعاد عالية الدقة، مما قد يؤثر على مجالات متعددة مثل الألعاب، والواقع الافتراضي، والتصميم.

Un nuevo estudio presenta el Modelo de Consistencia Latente Multiescala (MLPCM), que mejora la generación de formas 3D al aplicar modelos de consistencia avanzados de la síntesis de imágenes. Esta innovación es significativa ya que promete mejorar la eficiencia y calidad en la generación de formas 3D de alta resolución, lo que podría impactar en diversos campos como los videojuegos, la realidad virtual y el diseño.

Une nouvelle étude présente le Modèle de Cohérence Latente Multi-échelle (MLPCM), qui améliore la génération de formes 3D en appliquant des modèles de cohérence avancés issus de la synthèse d'images. Cette innovation est importante car elle promet d'améliorer l'efficacité et la qualité de la génération de formes 3D haute résolution, ce qui pourrait avoir un impact sur divers domaines tels que les jeux vidéo, la réalité virtuelle et le design.

A new study introduces the Multi-scale Latent Point Consistency Model (MLPCM), which enhances 3D shape generation by applying advanced consistency models from image synthesis. This innovation is significant as it promises to improve the efficiency and quality of generating high-resolution 3D shapes, potentially impacting various fields such as gaming, virtual reality, and design.

Multi-scale Latent Point Consistency Models for 3D Shape Generation

arXiv:2512.08506v1 Announce Type: new 
Abstract: A major challenge in reconstructing buildings from LiDAR point clouds lies in accurately capturing building surfaces under varying point densities and noise interference. To flexibly gather high-quality 3D profiles of the building in diverse resolution, we propose OCCDiff applying latent diffusion in the occupancy function space. Our OCCDiff combines a latent diffusion process with a function autoencoder architecture to generate continuous occupancy functions evaluable at arbitrary locations. Moreover, a point encoder is proposed to provide condition features to diffusion learning, constraint the final occupancy prediction for occupancy decoder, and insert multi-modal features for latent generation to latent encoder. To further enhance the model performance, a multi-task training strategy is employed, ensuring that the point encoder learns diverse and robust feature representations. Empirical results show that our method generates physically consistent samples with high fidelity to the target distribution and exhibits robustness to noisy data.

تم تقديم نموذج OCCDiff كنهج جديد لإعادة بناء هياكل المباني ثلاثية الأبعاد من سحب النقاط LiDAR المزعجة، باستخدام الانتشار الكامن في فضاء دالة الإشغال لتحسين دقة وجودة الملفات الشخصية ثلاثية الأبعاد المولدة. يتضمن هذا النموذج مشفر نقاط وهيكلية مشفر دالة لتسهيل توليد دوال الإشغال المستمرة عند دقات مختلفة.

Se ha presentado el modelo OCCDiff como un enfoque novedoso para reconstruir estructuras de edificios en 3D a partir de nubes de puntos LiDAR ruidosas, utilizando la difusión latente en el espacio de funciones de ocupación para mejorar la precisión y calidad de los perfiles 3D generados. Este modelo incorpora un codificador de puntos y una arquitectura de autoencoder de funciones para facilitar la generación de funciones de ocupación continuas a diversas resoluciones.

Le modèle OCCDiff a été introduit comme une nouvelle approche pour reconstruire des structures de bâtiments 3D à partir de nuages de points LiDAR bruyants, utilisant la diffusion latente dans l'espace de fonction d'occupation pour améliorer l'exactitude et la qualité des profils 3D générés. Ce modèle intègre un encodeur de points et une architecture d'autoencodeur de fonction pour faciliter la génération de fonctions d'occupation continues à diverses résolutions.

The OCCDiff model has been introduced as a novel approach to reconstructing 3D building structures from noisy LiDAR point clouds, utilizing latent diffusion in the occupancy function space to enhance the accuracy and quality of the generated 3D profiles. This model incorporates a point encoder and a function autoencoder architecture to facilitate continuous occupancy function generation at various resolutions.

Multi-scale Latent Point Consistency Models for 3D Shape Generation

Was this article worth reading? Share it

LucidQuery AI

Deptho.ai

Z3D