Deep Features and Extreme Learning Machines based Apparel Classification
Baris Gulbas, Abdulkadir Şengür, Emre Incel, Yaman Akbulut · 2019 International Artificial Intelligence and Data Processing Symposium (IDAP) · 2019
With the development of the new technologies, e-commerce, automated recommendation systems, smartphone applications have become inevitable tools of our lives. An apparel recommendation application highly attracts both women and men. An apparel recommendation application initially needs to determine the clothe types in a given image and then searches the internet for similar potential clothes. This task is quite challenging. With the development of the deep learning technologies, researchers were encouraged to propose solutions for such challenging problems. In this paper, features from pre-trained deep convolutional neural networks (CNN) and extreme learning machines (ELM) are used for apparel classification. To do it, AlexNet, VGGNet and ResNet models are considered. Fc6, fc7 and fc1000 layers of the pre-trained CNN models are used for feature extraction. These features are either used individually or concatenated form. These features are then classified with the ELM classifier. Classification accuracy is used to evaluate the achievements of the feature vectors. According to the obtained results, ResNet features obtain 60.42% accuracy score on apparel classification system (ACS) dataset. We also compare the obtained results with some of the published results on the same dataset. The comparisons show that our proposed scheme outperforms.