Data Driven SVBRDF Estimation Using Deep Embedded Clustering

Photo-realistic representation in user-specified view and lighting conditions is a challenging but high-demand technology the digital transformation of cultural heritages. Despite recent advances neural renderings, it still necessary to capture high-quality surface reflectance from photography controlled environment for real-time applications such as VR/AR arts. In this paper, we present deep embedding clustering network spatially-varying bidirectional distribution function (SVBRDF) estimation. Our designed simultaneously update basis its linear manifold spatial domain SVBRDF. We show that our dual scheme excels optimizing rendering loss terms convergence speed visual quality compared current iterative SVBRDF methods.