Micro-surface material reconstruction method and system based on deep learning

Abstract

The invention discloses a micro-surface material reconstruction method and system based on deep learning. The micro-surface material reconstruction method comprises the steps: acquiring a shot image of a real world material; inputting the shot image of the material into a pre-training generator network, and outputting a diffuse reflection map, a reflection map, a roughness map and a normal map which have higher resolution than the shot image; and drawing a rendered image by using the chartlet, wherein the appearance of the material presented by the rendered image is similar to that of the shotimage. A rendering image is drawn from a diffuse reflection map, a reflection map, a roughness map and a normal map by using a drawing module in a neural network on the basis of an adversarial generative network framework, and the rendering image and a real material shot picture are discriminated by using a discriminator module. Therefore, the defect that a traditional machine learning method needs to depend on a large number of material mapping labels is avoided, and the difficulty of material collection is reduced.

Description

technical field [0001] The present application relates to the technical field of image realistic rendering, in particular to a deep learning-based micro-surface material reconstruction method and system. Background technique [0002] The statements in this section merely mention the background art related to this application, and do not necessarily constitute the prior art. [0003] The material model mainly describes the local light reflection properties of the surface of the object, and the description of the material and accurate modeling play a vital role in the realism of the rendering result. Photorealistic material models are usually expressed as bidirectional reflectance distribution functions, and these functions can be divided into different models, among which the most commonly used model is the microsurface material model. Typical model parameters include diffuse reflection, reflection, roughness, normal, etc. [0004] Traditional material modeling works by sho...

AI Technical Summary

Problems solved by technology

[0005] The inventor found that the supervised deep learning method relies on a large training data set, and it is time-consuming and laborious to obtain such a large data set

In addition, due to fixed network parameters or limited video memory, texture maps are often low-resolution and cannot be used in practical applications.

Images