Hierarchical convolutional neural networks for fashion image classification

Deep learning can be applied in various business fields for better performance. Especially, fashion-related businesses have started to apply deep learning techniques on their e-commerce such as apparel recognition, apparel search and retrieval engine, and automatic product recommendation. The most important backbone of these applications is the image classification task. However, apparel classification can be difficult due to its various apparel properties, and complexity in the depth of categorization. In other words, multi-class apparel classification can be hard and ambiguous to separate among similar classes. Here, we find the need of image classification reflecting hierarchical structure of apparel categories. In most of the previous studies, hierarchy has not been considered in image classification when using Convolutional Neural Networks (CNN), and not even in fashion image classification using other methodologies. In this paper, we propose to apply Hierarchical Convolutional Neural Networks (HCNN) on apparel classification. This study has contribution in that this is the first trial to apply hierarchical classification of apparel using CNN and has significance in that the proposed model is a knowledge embedded classifier outputting hierarchical information. We implement HCNN using VGGNet on Fashion-MNIST dataset. Results have shown that when using HCNN model, the loss gets decreased and the accuracy gets improved than the base model without hierarchical structure. We conclude that HCNN brings better performance in classifying apparel.