Survey of Deep Learning Techniques for High-Fidelity Material Representation and Generation

High-fidelity material representation and efficient material generation techniques play a critical role in enhancing the rendering realism and appearance diversity for virtual scenes, and have long been a key research topic in computer graphics. Due to the complex optical properties and rich appearance variations of real-world materials, traditional approaches to material representation and generation still face significant challenges in terms of reconstruction quality, computational efficiency, and generation diversity. In recent years, the advancement of deep learning has introduced new paradigms for material representation and generation, demonstrating remarkable potential. This paper provides a systematic survey of recent research on high-fidelity material representation and generation based on deep learning. We first focus on several mainstream material models, including BRDF, SVBRDF, BTF, and BSSRDF, and provide an in-depth analysis of recent advances in neural material representation methods, along with a summary of related high-quality public datasets. Subsequently, we discuss key research directions in neural material generation, including material reconstruction from sparse observations, material generation under soft constraints, and the latest multimodal generation techniques. Finally, the paper outlines several promising future directions, such as the cooperation between neural representations and generative techniques, efficient and deployable neural material representation, and multimodal modeling with controllable generation. This survey aims to provide valuable references and insights for future research, and to promote the continued advancement and broader application of high-fidelity neural material representation and generation technologies.