Garment impression estimation with design parameters

Abstract

To estimate garment impressions, we verified three regression models using design parameters. Using three-dimensional apparel simulation, we generated 375 images of a men's outdoor jacket by changing design parameters: length, waist, hem circumference, and sleeve circumference. Nine people evaluated cool-uncool (kakkoī-kakkowarui in Japanese) impressions of the garment images using a semantic differential method. With the design parameters, we obtained the estimated image impression using three regression models: multiple linear regression (MLR), neural network (NN), and light gradient boosting machine (LightGBM). We used correlation coefficient(𝑐𝑜𝑟𝑟) and adjusted coefficient of determination between evaluated and estimated impression values to evaluate estimation performance. As a result, the LightGBM with the design parameters showed the highest mean 𝑐𝑜𝑟𝑟 for all participants. It was thus found that the design parameters are effective in estimating the garment impression with LightGBM.