arXiv:2510.03163v3 Announce Type: replace 
Abstract: We introduce ROGR, a novel approach that reconstructs a relightable 3D model of an object captured from multiple views, driven by a generative relighting model that simulates the effects of placing the object under novel environment illuminations. Our method samples the appearance of the object under multiple lighting environments, creating a dataset that is used to train a lighting-conditioned Neural Radiance Field (NeRF) that outputs the object's appearance under any input environmental lighting. The lighting-conditioned NeRF uses a novel dual-branch architecture to encode the general lighting effects and specularities separately. The optimized lighting-conditioned NeRF enables efficient feed-forward relighting under arbitrary environment maps without requiring per-illumination optimization or light transport simulation. We evaluate our approach on the established TensoIR and Stanford-ORB datasets, where it improves upon the state-of-the-art on most metrics, and showcase our approach on real-world object captures.

يقدم ROGR نهجًا جديدًا لإعادة بناء نماذج ثلاثية الأبعاد قابلة لإعادة الإضاءة للأجسام الملتقطة من زوايا متعددة، باستخدام نموذج إعادة إضاءة توليدي لمحاكاة تأثيرات الإضاءة البيئية المختلفة. تستخدم هذه الطريقة مجال إشعاع ثلاثي الأبعاد مشروط بالإضاءة (NeRF) الذي ينتج مظهر الجسم تحت أي إضاءة بيئية دون الحاجة إلى تحسينات مكثفة.

ROGR presenta un enfoque novedoso para reconstruir modelos 3D relightables de objetos capturados desde múltiples vistas, utilizando un modelo de relighting generativo que simula diversos tipos de iluminación ambiental. Este método emplea un Neural Radiance Field (NeRF) condicionado por la iluminación que permite obtener de manera eficiente la apariencia del objeto bajo cualquier condición de luz sin necesidad de optimización extensa.

ROGR introduit une nouvelle approche pour reconstruire des modèles 3D relightables d'objets capturés sous plusieurs angles, utilisant un modèle de relighting génératif pour simuler diverses illuminations environnementales. Cette méthode emploie un Neural Radiance Field (NeRF) conditionné par l'éclairage qui produit efficacement l'apparence de l'objet sous n'importe quelle condition d'éclairage sans optimisation extensive.

ROGR introduces a novel approach for reconstructing relightable 3D models of objects captured from multiple views, utilizing a generative relighting model to simulate various environmental illuminations. This method employs a lighting-conditioned Neural Radiance Field (NeRF) that efficiently outputs the object's appearance under any lighting condition without extensive optimization.

ROGR: Relightable 3D Objects using Generative Relighting

arXiv:2512.09335v2 Announce Type: replace 
Abstract: Modeling relightable and animatable human avatars from monocular video is a long-standing and challenging task. Recently, Neural Radiance Field (NeRF) and 3D Gaussian Splatting (3DGS) methods have been employed to reconstruct the avatars. However, they often produce unsatisfactory photo-realistic results because of insufficient geometrical details related to body motion, such as clothing wrinkles. In this paper, we propose a 3DGS-based human avatar modeling framework, termed as Relightable and Dynamic Gaussian Avatar (RnD-Avatar), that presents accurate pose-variant deformation for high-fidelity geometrical details. To achieve this, we introduce dynamic skinning weights that define the human avatar's articulation based on pose while also learning additional deformations induced by body motion. We also introduce a novel regularization to capture fine geometric details under sparse visual cues. Furthermore, we present a new multi-view dataset with varied lighting conditions to evaluate relight. Our framework enables realistic rendering of novel poses and views while supporting photo-realistic lighting effects under arbitrary lighting conditions. Our method achieves state-of-the-art performance in novel view synthesis, novel pose rendering, and relighting.

تم اقتراح إطار عمل جديد يسمى Relightable and Dynamic Gaussian Avatar (RnD-Avatar) لتحسين نمذجة الصور الرمزية البشرية القابلة لإعادة الإضاءة والتحريك من مقاطع الفيديو الأحادية، حيث يتناول التحديات المتعلقة بتحقيق نتائج واقعية من الناحية التصويرية بسبب نقص التفاصيل الهندسية المتعلقة بحركة الجسم. تستخدم هذه الطريقة أوزان جلد ديناميكية لتحديد تشوهات دقيقة تتغير حسب الوضع، كما تقدم تقنية تنظيم جديدة لالتقاط التفاصيل الهندسية الدقيقة.

Se ha propuesto un nuevo marco llamado Relightable and Dynamic Gaussian Avatar (RnD-Avatar) para mejorar la modelación de avatares humanos relightables y animables a partir de video monocular, abordando los desafíos para lograr resultados foto-realistas debido a detalles geométricos insuficientes relacionados con el movimiento del cuerpo. Este enfoque utiliza pesos de piel dinámicos para una deformación precisa que varía según la pose e introduce una nueva técnica de regularización para capturar detalles geométricos finos.

Un nouveau cadre appelé Relightable and Dynamic Gaussian Avatar (RnD-Avatar) a été proposé pour améliorer la modélisation des avatars humains relightables et animables à partir de vidéos monoculaires, en s'attaquant aux défis liés à l'obtention de résultats photo-réalistes en raison de détails géométriques insuffisants liés au mouvement du corps. Cette approche utilise des poids de peau dynamiques pour une déformation précise variant selon la pose et introduit une nouvelle technique de régularisation pour capturer des détails géométriques fins.

A new framework called Relightable and Dynamic Gaussian Avatar (RnD-Avatar) has been proposed to enhance the modeling of relightable and animatable human avatars from monocular video, addressing challenges in achieving photo-realistic results due to insufficient geometrical details related to body motion. This approach utilizes dynamic skinning weights for accurate pose-variant deformation and introduces a novel regularization technique for capturing fine geometric details.

ROGR: Relightable 3D Objects using Generative Relighting

Was this article worth reading? Share it

LucidQuery AI

MyArchitectAI

Deptho.ai

Uwear

4o Image Gen

Republiclabs.ai

Ready to build your own newsroom?