Data-Driven Shoe Last Generation Based on Preference-Aware Gan

Recent years have witnessed an increasing growth of customization in traditional computer-aided design and manufacturing, in which customized shoe last making has long been a pain to designers. Conventional last making solutions are not able to provide satisfactory comfortableness because they cannot take individuals' preferences into consideration. In this paper, we propose a data-driven approach for generating customized shoe lasts that provide satisfactory subjective comfortable experience for a particular group of people. First, we explore the correlation between the shape of a shoe last and a person's foot, and build a comfortableness prediction model to capture the correlation, with an accuracy rate up to 80%. Second, we design a 3D-last generation framework based on Wasserstein Generative Adversarial Networks (WGAN) [1]to generate customized 3D lasts that maximize the comfortableness expectation. Third, we implement our design in a real manufacturing environment and verify the effectiveness of generated lasts based on real-world foot shapes.