Clothing Classification Using Shallow Convolutional Neural Networks

Abstract

The online fashion industry is continuously growing with more requirements from applications such as 3D printing, digital clothing, and the Internet of Things to provide outstanding support for buyers. Many customers expect an algorithm capable of recognizing clothes may help clothing merchants better understand the profile of potential customers, focus sales targeting specific niches, design campaigns based on consumer preferences, and improve user experience. Therefore, artificial intelligence methods that detect and categorize human clothes are necessary to increase sales or better understand customers. Recent advancements in deep learning have sparked a slew of computer vision-based commercial applications. Object identification is used in various industries to speed up business processes where deep learning tools and techniques are leveraged in a variety of ways to improve service quality. One of them is clothing classification in the clothing business. The trained deep learning model can predict the name of any clothing by displaying an image of it, and this process can be repeated at a much quicker pace to tag thousands of items in a short amount of time with high accuracy. This study has examined shallow convolutional neural network architectures conducted from VGGNet architecture but in more compact forms with some stacks of layers, including convolutional, max pooling, and dropout layers. The proposed architecture has reached 0.9359 in classification accuracy on dataset Fashion-MNIST with 70000 samples, including images of the shirt, bag, and boot. The shallow convolutional neural networks architecture can outperform the classic machine learning method and better performance than VGGNet in classifying small-size images.