The type of fashion style that people prefer varies from person to person, and the classification of fashion styles often differs from person to person as well. Against this background, many methods have been proposed that generate different outfit images from a given outfit image. However, previous methods focus on the diversity and compatibility of the generated images, and often fail to reflect individual preferences. The purpose of this paper is to develop a system which outputs outfit images with partially modified outfit of the input image according to the user’s preferred style. The generation of an outfit image with some changes in the outfit of the input image is achieved by extracting and updating the features of the color, texture and shape of the clothes from the original image and its segmentation mask. To classify images into the style of the individual’s preference, we create the user’s original dataset to learn the user’s preference in advance by asking the user to classify a group of images into 4 styles. We conducted an evaluation experiment of our system to confirm that our system reflects user’s individual preference. As a result of the evaluation experiment, it was confirmed that the same image was recommended as different styles for different users, and that the users also thought that the recommended style was match to the style that the user’s classification of the style. There are some challenges related to the bias of the data used and the learned model used to generate the images. Therefore, it is expected that improving these points will result in a system with higher recommended accuracy.
Paper
Full text
Fashion Recommendation System Reflecting Individual’s Preferred Style
OpenAlex · Generative Adversarial Networks and Image Synthesis · 2021
Abstract
The type of fashion style that people prefer varies from person to person, and the classification of fashion styles often differs from person to person as well. Against this background, many methods have been proposed that generate different outfit images from a given outfit image. However, previous methods focus on the diversity and compatibility of the generated images, and often fail to reflect individual preferences. The purpose of this paper is to develop a system which outputs outfit images with partially modified outfit of the input image according to the user’s preferred style. The generation of an outfit image with some changes in the outfit of the input image is achieved by extracting and updating the features of the color, texture and shape of the clothes from the original image and its segmentation mask. To classify images into the style of the individual’s preference, we create the user’s original dataset to learn the user’s preference in advance by asking the user to classify a group of images into 4 styles. We conducted an evaluation experiment of our system to confirm that our system reflects user’s individual preference. As a result of the evaluation experiment, it was confirmed that the same image was recommended as different styles for different users, and that the users also thought that the recommended style was match to the style that the user’s classification of the style. There are some challenges related to the bias of the data used and the learned model used to generate the images. Therefore, it is expected that improving these points will result in a system with higher recommended accuracy.