Every human being has an individual shape. In view of the sheer variety of body shapes, it is not always easy to find clothes that are not only fashionable, but also “fit” perfectly. But traditional clothing sizes often reach their limits when it comes to the right fit, which is why experts from the University of Stuttgart and the German Institutes for Textile and Fiber Research Denkendorf (DITF) are using AI to help classify body shapes and to find the right measure.
Off-the-shelf sizes leave plenty of scope for a redesign. The calculation is based on standard tables based on body measurements - usually body size as well as chest and waist circumference, but there is no binding standard and each fashion label has its own contour styling. So the same item of clothing will not necessarily fit two people of the same size and the same individual can rarely wear the same size in all garments.
“Simply measuring the body,” Prof. Meike Tilebein explains, “does not enable us to completely reproduce the overall body shape, which, for example, is also determined by the relational proportions between individual parts of the body.” The cyberneticist heads the Institute for Diversity Studies in Engineering (IDS) at the University of Stuttgart as well as the Center for Management Research at the DITF. She has carried out research into diversity in many of its varied facets and she and her team have analyzed the diversity of body shapes in collaboration with Avalution GmbH, a provider of services and software for fashion and body models. They have now succeeded in improving the fit of clothing on the basis of morphotype classification and with the aid of AI methods.
A look at the body scans of various people, all of whom take a size 38, but each with a different body shape, shows how wide the range can be. Hollow back or upright posture? Knock-kneed or strong thighs? Square or sloping shoulders? “These and many other morphological features contribute to whether or not clothes fit properly,” explains Thomas Fischer, a scientist at the DITF. An analysis of data from the “SizeGermany” series measurement pool compiled by Avalution illustrates the problem: of 455 representative women's bodies, 331 could not be assigned an unambiguous size.
The team initially developed ten basic features for the new classification model, which are subdivided into the entire body and the upper and lower bodies. These include, for example, typical basic shapes such as the “triangle” with narrower shoulders than hips or the “rectangle”, in which shoulders and hips have a similar width. Sorting is also done according to the waist shape: is the waist of the clearly pronounced hourglass type or more like the circumference of a sphere? In the case of the lower body, the model also takes account of the leg length, and in the case of the upper body, the ratio between chest circumference and back length.
Four human experts independently compared each of these ten characteristics with the 455 body scans from the “SizeGermany” database. The result is a validated knowledge base that is available to the AI algorithms as a training set and, above all, is scalable. “This offers the possibility of classifying much larger data pools with the help of AI,” says Tilebein.
A long-established AI sub-area called Case Based Reasoning (CBR) is used for this purpose. This method of machine learning imitates the behavior of human professionals and is based on the principle of similarity, i.e., just like humans, the software learns and works under the premise that similar solutions exist for similar problems.
Benefits for manufacturers and customers
Applied to the classification of body shapes, the process works like this: the algorithms developed at the DITF keep the so-called CBR cycle going. If a new person or figure is to be classified, the algorithm uses the “old cases” stored in the case base and searches for the classification that is most similar to the new case. Based on this template, the system develops a classification proposal. This is adjusted as far as possible and then saved as a new classification in the case base. In this way, the AI engine expands its knowledge incrementally.
The CBR method not only allows conclusions to be drawn about the most suitable individual size, but also, for example, about the frequency and distribution of certain morphotypes, which offers enormous benefits both to fashion companies, but also their customers. Companies can better analyze the market and optimize their products by using more precisely fitting ready-made sizes and plan production in a more targeted manner. Particularly in e-commerce, providers can recommend the individually best product in the optimum size and thus minimize unnecessary returns. And the customers should also be satisfied with their new favorite piece for a longer period, if it fits perfectly.
The new technology is now being used successfully in practice. Morphotypes are included in Avalution's analyses and the analyses carried out for special customer groups at the DITF were also successful. Tilebein and her team are currently investigating the application potential in online retailing and are looking into whether and to what extent artificial neural networks are superior to the CBR method. So will AI introduce more affordable individualization to the fashion industry? “We already have the vision of using digital technologies to get as close as possible to the individual 'fit' before we start sewing,” says Tilebein. “But whether this will work in mass production scenario remains to be seen.”
Dr. Jutta Witte