ITR: Collaborative Research: Generating an Accurate Sense of Depth and Size Using Computer Graphics

尽管在过去的十年里，计算机图形学在现实主义方面取得了令人印象深刻的进步，但目前还无法有效地生成看起来很大的物体和环境的图像。这主要是因为计算机图形在传达绝对深度信息方面表现不佳。这个项目的目标是证明，如果渲染方法的开发特别注意传达绝对深度线索的需要，就有可能显著改善计算机图形的深度感和比例感。实现这一目标将需要对从2D图像中提取的3D信息有新的见解，修改图形算法以更好地呈现突出信息，以及复杂的感知实验来验证人们实际上可以看到预期的3D空间。PI的方法将利用计算视觉的结果和方法，这在计算机图形学社区中是前所未有的。计算视觉提供了关于如何从2D图像中恢复3D空间信息的内在约束的见解。特别是视觉的计算分析指出了相对深度判断和绝对深度判断的重要区别。令人惊讶的是，通常研究的图像线索实际上很少足以提供绝对深度的信息。由于显示技术的基本限制，以及我们无法精确控制除非在沉浸式环境中观看的条件，在那些可以做到的技术中，有一些不能在计算机图形学中利用。该研究将影响广泛的图形应用，其中需要传达准确的空间信息，包括教育和培训，设计和原型，以及远程呈现。

Despite impressive gains in realism over the last decade, computer graphics is currently unable to effectively generate images of objects and environments that look large. This is mostly because computer graphics is poor at conveying information about absolute depth. The goal of this project is to demonstrate that it is possible to significantly improve the sense of depth and scale in computer graphics if rendering methods are developed with specific attention to the need to convey cues for absolute depth. Accomplishing this goal will require new insights into the 3D information extractable from 2D images, modifications to graphics algorithms in order to better render salient information, and sophisticated perceptual experimentation to validate that people can actually see the intended 3D space. The PI's approach will be to draw upon the results and methods of computational vision in ways that have not previously been done in the computer graphics community. Computational vision provides insights into the intrinsic constraints on how information about 3D space can be recovered from 2D images. In particular, the computational analysis of vision points out the important distinction between relative depth judgments and absolute depth judgments. Surprisingly few of the commonly studied image cues are in fact sufficient to provide information about absolute depth. Of those that do, several cannot be exploited in computer graphics due to fundamental limitations in display technology and our inability to precisely control viewing conditions except in immersive environments. The research will impact a broad range of graphics applications in which accurate spatial information needs to be conveyed, including education and training, design and prototyping, and telepresence.