EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

MIKU: Material Invariance and Knowledge Unification for PBR Generation

Yajie Bao, Jiaye Fu, Youtian Lin, Shuang Wu, Jiachen Qian, Feihu Zhang, Xun Cao, Yao Yao

ACM Transactions on Graphics · 09.11.2026 · 45 (6) · 1-17

DOI: 10.1145/3842521 ↗

Kurz gefasst

Fachartikel in der Zeitschrift ACM Transactions on Graphics, online erschienen am 09.11.2026. Verfasst von Yajie Bao und 7 weiteren Autor:innen. Die Arbeit nennt 72 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, Gusu Innovation & Entrepreneurship Leading Talents Program.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In physically based rendering (PBR) content creation, high-quality material generation remains a fundamental challenge, primarily hindered by the absence of standardized representations across existing datasets and synthesis pipelines. The heterogeneity in material definitions, rendering conventions, and parameterization schemes introduces significant inconsistencies, ultimately leading to unstable and physically implausible PBR synthesis results. To systematically address these limitations, we propose MIKU, a unified framework that establishes standardized PBR asset representation and enables physically consistent material generation across diverse 3D content domains. Our core contribution is a comprehensive unified PBR material processing pipeline that standardizes the entire workflow from data preprocessing to final texture synthesis. This pipeline enforces consistent representation across geometry and texture domains throughout material generation. Critically, it addresses the long-standing challenge of inconsistent material parameterizations by introducing a normalization scheme that harmonizes diverse input sources into a common representation space, enabling robust cross-dataset training and inference. Building upon this principled formulation, we train a novel physically aware diffusion model specifically designed for PBR texture generation. Our model operates within a two-stage multi-view synthesis framework. The first stage focuses on geometry-aware albedo generation, where multi-view albedo maps are synthesized under explicit geometric guidance to ensure cross-view appearance consistency. The second stage performs metallic-roughness (MR) generation using our trained diffusion model, which predicts physically plausible roughness and metallic maps conditioned on the generated albedo and the unified representation. The model is trained end-to-end on standardized data, learning to generate materials that respect category-level material priors and semantic part boundaries. By explicitly decoupling representation standardization from appearance synthesis, MIKU substantially improves both the stability and physical consistency of generated materials. Extensive experiments demonstrate that our approach achieves state-of-the-art performance in both texture quality and semantic consistency across different object parts, significantly advancing automated material generation for 3D content creation workflows.

Förderung

  • National Natural Science Foundation of China
  • Gusu Innovation & Entrepreneurship Leading Talents Program

Zitieren

APA

Bao, Y., Fu, J., Lin, Y., Wu, S., Qian, J., Zhang, F., Cao, X., & Yao, Y. (2026). MIKU: Material Invariance and Knowledge Unification for PBR Generation. ACM Transactions on Graphics, 45(6), 1-17. https://doi.org/10.1145/3842521

Vancouver

Bao Y, Fu J, Lin Y, Wu S, Qian J, Zhang F, et al. MIKU: Material Invariance and Knowledge Unification for PBR Generation. ACM Transactions on Graphics. 2026;45(6):1-17. doi:10.1145/3842521

BibTeX
@article{Bao2026,
  author = {Bao, Yajie and Fu, Jiaye and Lin, Youtian and Wu, Shuang and Qian, Jiachen and Zhang, Feihu and Cao, Xun and Yao, Yao},
  title = {MIKU: Material Invariance and Knowledge Unification for PBR Generation},
  journal = {ACM Transactions on Graphics},
  year = {2026},
  volume = {45},
  number = {6},
  pages = {1--17},
  doi = {10.1145/3842521}
}
Kostenlos starten →