| dc.description.abstract | Modern computerized weft knitting machines enable on-demand production of custom, whole garments at once. They reduce the need for manual post-processing and generate minimal waste. Yet, their programming is still hardly accessible and is effectively done manually by few skillful knitting technicians. The programming of knitted garments typically involves scheduling hundreds of thousands of stitches. While every individual stitch created on such machines can, in theory, be controlled digitally, the ability to effectively do so depends heavily on the programming software being sufficiently accessible to the user. Unfortunately, current knitting software is typically closed and relies mostly on low-level programming. The lack of standardization and more accessible, higher-level user design tools effectively hinder the possibility of a digital, on-demand production of garments for all.
In this thesis, I explore the design space that flat-bed, weft knitting machines span and propose novel design workflows to enable accessible, digital customization of garments created on these machines. First, I introduce the inverse design problem of automatic knitting program generation from a single image, together with a machine learning framework that enables it. Second, I describe a parametric, primitive-based design tool that merges inspirations from both computer-aided design and pixel-based image editing. Finally, I propose a novel workflow to translate traditional, sketch-based garment patterns into knitting programs. The resulting system allows anyone to harness the plethora of existing garment designs while providing knitting-specific customization capabilities. | |
