Sculpt and sketch interfaces for 3D modeling

点击翻译按钮获取中文摘要

Sketching and sculpting have been the medium of visual communication for millenia. Since the early days of computing, as far back as Ivan Sutherlannd's seminal Sketchpad system (1963), interfaces based on a sketch or sculpt metaphor have been touted as "natural" approaches to 3D modeling. The last two decades have seen a number of promising sketch or sculpt based 3D modeling systems but the central goal of transforming a visual idea into a digital 3D model continues to be a fertile area of ongoing research. In the past 5 years my research has culminated in successful 3D modeling systems: ILoveSketch and Analytic 3D drawing (sketch-based), Shapeshop3D and MeshMixer (sculpt-based). This and similar research in computer graphics has been largely oblivious of an important body of research in human visual perception that studies the various biases inherent limitations in human perception and understanding of 3D shape. This proposal thus intends to explore sculpting and sketching for 3D shape modeling with greater rigour by first attempting to computationally model the various biases we have in our perception of 3D shape. For example, it is well-known that when asked to draw a cube from an oblique view, humans elongate it in depth significantly. Without a model for this bias, a computer would erroneously infer the drawing to be an elongated cuboid. Conversely, a digital cube, if displayed without correcting for our visual bias, would appear to us as a squashed cuboid. A significant aspect of the proposed research will thus build upon prior research in human perception, shape understanding, abstraction and aesthetics in art and test and model various hypotheses relating to biases in visual perception. Despite the existence of 3D input devices and displays, the bulk of input devices and displays in use today continue to be 2D. Even with a good model of the biases in human perecption the ambiguities in inferring 3D shape from 2D input is what makes digital 3D modeling a uniquely challenging problem. There are two ways to address this fundamental problem: by building better inference engines that use context and likelyhood models of 3D shape to resolve ambiguities (eg. Analytic 3D drawing) or by building better interfaces and interactive workflows that allow users to augment their 2D input and explicitly express their intended 3D shape (eg. ILoveSketch). This research will continue to explore both these research directions.