A Method to Obtain Effective Attributes for Attractive Cosmetic Bottles by Deep Learning

Abstract

Kansei values are critical factors for manufacturing. Kawaii, an adjective that denotes such positive meanings as cute or lovable, has become even more critical as a kansei value. Therefore, we believe that kawaii will be a key factor in future product design. We focus on systematic study for kawaii products in terms of such physical attributes as color, shape, and material perception. In this research, we proposed a method to obtain effective attributes for kawaii cosmetic bottles. First, we collected evaluation results of the kawaii of cosmetic bottle images and constructed a model using deep learning. Then we modified the images focusing on particular attributes and evaluated them using our model to obtain effective attributes. Finally, we verified those attributes by experimentally evaluating the modified images. The novelty of this research is our new method to obtain effective attributes, which can be applied to other products and other kansei values.