Deep Learning Based Image Segmantation and Classification for Fashion Detection on Smartphones

With the rapid increase in smartphone technologies and social media applications, photo sharing among people has become widespread. Most of the photos shared include the clothes people use in daily life. Access to desired clothes is easier with online facilities. Identifying the outfit in a photograph is the first step to this. In this study, an original system with deep learning based segmentation and classification methods aiming at detecting the clothes used in daily life via smart phone is proposed. The image taken with a smartphone camera was segmented first, and the clothing boundaries in the image were determined and tagged for classification. The labeled image was trained with the convolutional neural network and the category to which the garment belongs was determined. Using different models in the segmentation part, the highest accuracy was obtained in the mask regional convolutional neural network on the DeepFashion2 da taset. In the classification part, the Xception architecture showed superiority in the trainings made with different convolutional neural networks. The models with the highest accuracy rate are combined with our Android application called Clothes Detection, which we developed to detect clothes. Thanks to our user-friendly application that has been developed, clothes can be detected on the photograph.