Object detection and semantic segmentation of fashion images

Abstract

Over the past few years, fashion brands have been rapidly implementing computer vision into the fashion industry. Our research objective was to analyse a number of methods suitable for object detection and segmentation of apparel in fashion images. Two types of models are proposed. The first, simpler, is a convolutional neural network used for object detection of clothing items on the Fashion-MNIST dataset and the second, more complex Mask R-CNN model is used for object detection and instance segmentation on the iMaterialist dataset. The performance of the first proposed model reached 93% accuracy. Furthermore, the results from the Mask R-CNN model are visualized.