arXiv:2504.03536v2 Announce Type: replace 
Abstract: Single-image human reconstruction is vital for digital human modeling applications but remains an extremely challenging task. Current approaches rely on generative models to synthesize multi-view images for subsequent 3D reconstruction and animation. However, directly generating multiple views from a single human image suffers from geometric inconsistencies, resulting in issues like fragmented or blurred limbs in the reconstructed models. To tackle these limitations, we introduce \textbf{HumanDreamer-X}, a novel framework that integrates multi-view human generation and reconstruction into a unified pipeline, which significantly enhances the geometric consistency and visual fidelity of the reconstructed 3D models. In this framework, 3D Gaussian Splatting serves as an explicit 3D representation to provide initial geometry and appearance priority. Building upon this foundation, \textbf{HumanFixer} is trained to restore 3DGS renderings, which guarantee photorealistic results. Furthermore, we delve into the inherent challenges associated with attention mechanisms in multi-view human generation, and propose an attention modulation strategy that effectively enhances geometric details identity consistency across multi-view. Experimental results demonstrate that our approach markedly improves generation and reconstruction PSNR quality metrics by 16.45% and 12.65%, respectively, achieving a PSNR of up to 25.62 dB, while also showing generalization capabilities on in-the-wild data and applicability to various human reconstruction backbone models.

تقدم HumanDreamer-X تقدمًا كبيرًا في إعادة بناء الإنسان من صورة واحدة، حيث تتناول التحديات المتعلقة بالتناسق الهندسي والواقعية البصرية. يدمج هذا الإطار الجديد بين توليد وجهات متعددة وإعادة البناء، محققًا تحسينًا في PSNR بنسبة 16.45% وPSNR نهائي يبلغ 25.62 ديسيبل، مما يعزز جودة النماذج ثلاثية الأبعاد المستخدمة في تطبيقات نمذجة الإنسان الرقمية.

La introducción de HumanDreamer-X representa un avance significativo en la reconstrucción humana a partir de una sola imagen, abordando desafíos en la consistencia geométrica y la fidelidad visual. Este nuevo marco integra la generación de múltiples vistas y la reconstrucción, logrando una mejora del PSNR del 16,45% y un PSNR final de 25,62 dB, lo que mejora la calidad de los modelos 3D utilizados en aplicaciones de modelado humano digital.

L'introduction de HumanDreamer-X représente une avancée significative dans la reconstruction humaine à partir d'une seule image, en abordant les défis de la cohérence géométrique et de la fidélité visuelle. Ce nouveau cadre intègre la génération multi-vues et la reconstruction, atteignant une amélioration du PSNR de 16,45 % et un PSNR final de 25,62 dB, ce qui améliore la qualité des modèles 3D utilisés dans les applications de modélisation humaine numérique.

The introduction of HumanDreamer-X marks a significant advancement in single-image human reconstruction, addressing challenges in geometric consistency and visual fidelity. This novel framework integrates multi-view generation and reconstruction, achieving a PSNR improvement of 16.45% and a final PSNR of 25.62 dB, which enhances the quality of 3D models used in digital human modeling applications.

HumanDreamer-X: Photorealistic Single-image Human Avatars Reconstruction via Gaussian Restoration

Was this article worth reading? Share it

Ready to build your own newsroom?