DreamyMirror A Machine Learning-Based Personalized Outfit Recommendation System Integrating Skin Tone Classification and Size

Abstract

Due to the increasing popularity of e-commerce, buying has undergone a tremendous shift and belongs to a different world. ” But there is still some difficulties with personalization when it comes to the proposed size and the choice of the color taking into consideration the skin color of the person. In this paper, the authors introduce DreamyMirror, a novel system for recommending outfits based on the individual user’s skin tone and size, along with allowing for virtual fitting. Based on a skin tone classifier CNN model built from CelebA database and a pre-trained Pose_iter_440000 of OpenPose. size prediction caffemodel, DreamyMirror provides unique suggestions of outfits according to the appearance of the person. The system also contains the Virtual Try on system wherein users can anticipate what the selected apparels look like at a glance. The proposed system can automatically recognize skin tone with an overall classification accuracy of 92% and the size prediction accuracy is found to be 95%. Moreover, the user engagement metrics also show an 85% level of satisfaction with the features – this is about the virtual try-on. This work shows how the application of AI in fashion retail can help better engage the customer through a truly individualized shopping experience and, at the same time, minimize the return rate because of the wrong size or color choice.