CloNet: A Fine-Grained Classification Network for Clothing Style Image

Abstract

As the most fundamental application in the field of fashion, clothing category classification has attracted the attention of more and more scholars. The diversity of clothing styles leads to a complex process of clothing category classification. To accurately and quickly distinguish clothing categories, we propose a new clothing classification algorithm, namely CloNet. The key ideas of the framework are as follow: 1) the inverted residual structure is used to fully extract the clothing texture features, 2) dilated convolution with different dilation rates is used to further learn semantic features of garments, 3) the width factor is used to find the optimal network structure, and 4) a new classification network is proposed to automatically learn clothing features in a lightweight and efficient process. Comprehensive and rich experiments demonstrate our discoveries and the effectiveness of our model. We improve the top-1 classification accuracy by 0.8% compared to the state-of-the-art models and the model size is only one-third of its size.