Drawings are widely used in the object design process. Beyond form, these drawings are also used to represent the material on the object's surface (e.g. polished metal or shiny plastic). However, the “correct” and accurate representation of materials in these drawings is a challenge for many designers. In this project, we propose an approach to partially automate the creation of such drawings from a shape drawing and a target material.
These representations are generally not realistic, but are stylized in the drawing process and can have different levels of abstraction. In this thesis, we will first study material sketches in design to better understand how humans perceive them and how they are constructed. We will then use this knowledge to generate synthetic drawings and train a deep learning algorithm to interpret them, i.e. extract material information from a designer's drawing. These tools will speed up the design process by enabling designers' drawings to be reused and augmented.