Single Image Clothing Style Recognition Using Mask RCNN

Abstract

Visual fashion style recognition from social media is a key application for online marketing. Currently, such is achieved fully manually, which is ineffectively. We aim to solve it using artificial intelligence. In this paper, a clothing style recognition algorithm based on Mask RCNN is proposed. As the most advanced convolutional neural network, Mask RCNN has made great progress in all aspects of image processing. The paper also contributes a clothing dataset with five different styles. After iterative training, the final MAP of the model reaches 0.79, which can meet the requirements of high-precision clothing recognition. To enable deep learning training and testing, we also propose a fashion style dataset. The experimental results also demonstrate the excellent performance of Mask RCNN in clothing style recognition.