Abstract: Realistic rendering of large-scale complex 3D scenes is one of the key technologies in the fields of 3D animation, digital humans, virtual reality, virtual simulation, digital twins, video games, etc. Surface materials with a high degree of realism are the cornerstones of 3D rendering, and the accuracy and fineness of their parametric maps determine the degree of realism of the rendering results. In this paper, according to the different equipment and material representations, the related work is classified into three categories: precision equipment measurement, procedural node graph material creation, and lightweight equipment-based recovery. The evaluation indexes and capabilities of surface material inverse recovery methods are also summarized. Firstly, this paper introduces the measurement methods based on precision equipment and analyzes the correlation between professional equipment and the development of material representation models. Then, material reconstruction and generation methods based on procedural node graphs are briefly described, the characteristics of related software are summarized, and the development status and problems of procedural material inversion methods based on style loss optimization are highlighted. In addition, this paper reviews the efficient reconstruction and generation of materials based on lightweight equipment, describes the criteria for lightweight equipment acquisition, and compares and analyzes the advantages and disadvantages of various deep learning methods in terms of extracting latent spatial features of materials, estimating self-similarity of materials, suppressing the effect of highlights on material recovering, designing self-supervised training of materials, and generating materials by multimodal guided generation. Finally, the state of the art in inverse recovery of highly realistic surface materials is summarized. This paper pointed out that the current surface material recovery task still faces the problems of harsh acquisition conditions, difficulties in recovering complex materials such as clear coatings and glints, high cost of network training, and susceptibility to strong highlights. Future work will focus on the modeling of special classes of materials, soft-mask prediction, and region redrawing for strong highlights, material extraction from 3D objects, and accurate material editing.