Predicting User’s Measurements without Manual Measuring

Abstract

Featured Application: Improving garment selection without manual measuring for made-to-order sports garments. As sports garments are stretchable, different sizing tables are used than for retail clothing. However, customers measuring themselves leads to errors and unsatisfaction, since these customized branded garments cannot be returned. Using fitting sets avoids this, but this is not always feasible, especially in an online retail environment. Therefore, this research aims to use descriptive measures—parameters that do not require manual measuring because they are readily known by heart by almost any customer—to predict users’ body measurements, which can, thus, be used by customers to determine the size of their sports garment from a sizing chart. To validate if these input measures are sufficient to predict the correct size, three prediction methods are used and compared with baseline manual measurements. The methods are: (i) clothing size predictions from shape models with descriptive measures as inputs, (ii) clothing size predictions from a regression analysis, and (iii) clothing size predictions from a shape model based on extensive 3D scanned measurements as input. The conclusion is that a regression algorithm with, as input variables, the straightforward demographics of age, gender, stature, and weight is more accurate than the algorithm with the same inputs but with a shape model behind it. Moreover, chest and hip circumferences have an intraclass correlation coefficient rating above 0.9 and are, thus, suited for online retail of stretchable garments, such as cycling clothes. As validated by end-users, the regression predictions are shown to agree with preferred garment sizes of the participants, within the natural variation of personal preferences.