Deep Shading: Convolutional Neural Networks for Screen Space Shading

被引：88

作者：

Nalbach, O. ^{[1
]}

Arabadzhiyska, E. ^{[1
,2
]}

Mehta, D. ^{[1
]}

Seidel, H. -P. ^{[1
]}

Ritschel, T. ^{[3
]}

机构：

[1] Max Planck Inst Informat, Saarbrucken, Germany

[2] Saarland Univ, MMCI, Saarbrucken, Germany

[3] UCL, London, England

来源：

COMPUTER GRAPHICS FORUM | 2017年 / 36卷 / 04期

关键词：

CCS Concepts; Rasterization; Rendering; •Computing methodologies → Neural networks;

D O I：

10.1111/cgf.13225

中图分类号：

TP31 [计算机软件];

学科分类号：

081202 ; 0835 ;

摘要：

In computer vision, convolutional neural networks (CNNs) achieve unprecedented performance for inverse problems where RGB pixel appearance is mapped to attributes such as positions, normals or reflectance. In computer graphics, screen space shading has boosted the quality of real-time rendering, converting the same kind of attributes of a virtual scene back to appearance, enabling effects like ambient occlusion, indirect light, scattering and many more. In this paper we consider the diagonal problem: synthesizing appearance from given per-pixel attributes using a CNN. The resulting Deep Shading renders screen space effects at competitive quality and speed while not being programmed by human experts but learned from example images.

引用

页码：65 / 78

页数：14