Single Image Intrinsic Decomposition with Discriminative Feature Encoding

Zongji Wang, Feng Lu

Abstract: Intrinsic image decomposition is an important and long- standing computer vision problem. Given a single input im- age, recovering the physical scene properties is ill-posed. In this work, we take the advantage of deep learning, which is proven to be highly efficient in solving the challenging com- puter vision problems including intrinsic image decomposi- tion. Our focus lies in the feature encoding phase to extract discriminative features for different intrinsic layers from a single input image. To achieve this goal, we explore the distinctive characteristics between different intrinsic com- ponents in the high dimensional feature embedding space. We propose a feature divergence loss to force their high- dimensional embedding feature vectors to be separated ef- ficiently. The feature distributions are also constrained to fit the real ones. In addition, we provide an approach to re- move the data inconsistency in the MPI Sintel dataset, mak- ing it more proper for intrinsic image decomposition. Ex- perimental results indicate that the proposed network struc- ture is able to outperform the state-of-the-art methods.