Creating Virtual Shapes via Intuitive Input

点击翻译按钮获取中文摘要

The manufacturing and the entertainment industries are heavily automated but still far from being efficient. One of the key issues is creating content. Engineers and artists spend days and months designing digital CAD models to be manufactured or designing and animating character models for videogames and special effects. As a result, content creation processes are often the most time consuming and expensive processes in those industries. The main reason for such hindrance is shape modeling software. Current shape modeling systems for animation or CAD are built with a tool-oriented paradigm in mind. In this paradigm, users think of modeling as a sequence of operations with tools: for instance, draw a rectangle, create a curve, extrude along it, smooth. However, when a person wants to communicate their idea of a shape to another person, such tool-oriented explanations would be excessively time-consuming. This gap between the way we transmit information to each other versus the way we transmit information to a computer is a fundamental problem of modern geometric modeling tools. Despite significant progress, current shape modeling software requires specialized training and often hampers artists' imagination by forcing them to think about details instead of the big picture. The goal of my research program is to create modeling software that is human-oriented instead of tool-oriented. This kind of software is designed to infer a base shape given a very concise, natural, yet often ambiguous description. Such software will unleash artists' creativity and productivity. Expanding the current modeling toolkit with human-oriented software will not only speed up prototyping of 3D models, but also enable artists with little specialized 3D training to create high-quality models. In the long term, I am envisioning shape modeling software for various industries and applications, where traditional computer modeling tools coexist with human-oriented tools based on natural inputs. This effort will eventually reshape our way of thinking about modeling shapes: from a tedious process to an effortless recording of one's imagination. In this proposal, I will focus on sketches and gesture drawings as the main forms of natural inputs. Initial sketches can be astonishingly close to the final rendered image of a posed 3D model yet are still used only as a reference in industrial modeling pipelines. In contrast, I focus on using sketches directly as an input to the modeling/animation system, aiming to sidestep the tedious manual modeling or posing. While we are far from algorithmically interpreting natural sketches and gesture drawings, we can come significantly closer to the goal with the proposed projects.