P-23

Nowadays, many clothing companies use webshops alongside their physical retail outlets. However, a downside of webshops is that shoppers cannot fit clothes; this requires them to visit the physical stores. The project 'Fitting Fashion through the Internet' ('Passende Mode via Internet') tries to overcome this issue by investigating technologies that assist shoppers in choosing correctly fitting clothing. At Saxion, the research group Ambient Intelligence focuses on enabling this through the use of machine learning. We are working towards an automated application which can estimate a person’s body measurements based on 2D photographs of that person. To a large extent, this research is carried out by Software Engineering students.

Machine learning is an application of artificial intelligence in which algorithms use existing data to 'learn' to classify or estimate new data. In our project 'Fitting Fashion Using Machine Learning' (FFML), we train such algorithms based on a data set consisting of 2D photographs of people and corresponding body measurements determined by tailors. This is done by using convolutional neural networks: a class of deep learning neural networks which is dedicated to analyzing visual information.

Our FFML system makes use of these neural networks to generate a model which determines 14 distinct body measurements used for fitting clothing. It requires two photographs as input to estimate a person’s measurements: a person's side- and front-view. Currently, the FFML system can estimate a person's measurements with, on average, a 2 cm deviation.

We expect the FFML system to perform even better as we are iteratively approaching the problem from an increasingly focused perspective. For example, we are designing models which are apparel-specific: for shirts, pants, etc. Additionally, we are making the system more user-friendly by providing users with an intuitive front-end interface through which they can upload photographs and have their sizes estimated.