Computer vision and machine learning for the creative industry

Abstract

Computer Vision is becoming a prominent tool for the creative industry. Fashion can clearly benefit from automation techniques and the processing power of modern processors applied to Machine Learning and Machine Vision.

In this thesis we will explore three years of collaboration of the author with Adidas, one of the leaders in the sport fashion market, and the systems and research topics developed as consequence. These systems are related to real, specific needs of a company as big as Adidas, while still providing high profile research material.

The first research topic is Computer Vision as a Graphics tool. Every creative company has designers, each one uses its own tools, but each creates proof of concepts and designs that need to be converted in final products and renderings. This process involves sketch vectorization, i.e. transforming a raw paper sketch in a vectorized collection of B-splines. Computer vision can perform this task automatically, saving much of designers' time. Part of this thesis will discuss this topic and show how we have beaten the state of the art in this field, proposing a new line extraction method, a new thinning algorithm and a complete system to perform such vectorization.

The second topic of the thesis will be Computer Vision and Machine Learning as tools for analysis. Medium-to-large fashion companies will sooner or later find the need of analyzing their production during time and look at competitors. Adidas is no exception, since it has development teams across the world and a huge amount of new products per year. Moreover, fashion has seasons and trends that rapidly change, and it is of paramount importance to monitor constantly what emerges as a tendency, and what is “old” and needs to be ceased. A way to do this is analyzing thousands of products pictures and relating them to the sales. To do this, we need tools for image analysis, in the form of Computer Vision (color palette estimation, template matching for detection), and Machine Learning (classification, clustering). This thesis will discuss a complete treatment of the problem and propose a comprehensive system for analysis and extraction of these features. Experimental results will show its optimal performance compared to actual human work.

The thesis will also briefly introduce Deep Learning as a tool for creation, creativity and recommendations. A system to transform raw shoe sketches to colored renderings will be introduced. Another system able to learn style transfer and product altering will also be proposed.