CA-FashionNet: Light-Weight Hybird Model for Fashion Style Classification

A light-weight fashion style classification model is crucial for the online fashion industry. However, the varied and subtle differences between styles make extracting and learning distinctive features challenging. To address this issue, we propose a light-weight hybrid network named CA-FashionN et, which combines convolution and attention mechanism to enhance the effectiveness of feature extraction. First, CAFashionNet constructs the group extraction blocks to effectively capture fine-grained local features in fashion styles. Then, related attention blocks are introduced into CA-FashionNet to capture the internal correlations in high-level fashion features. Our proposed CA-FashionNet is evaluated extensively on the FashionStyle14 benchmark dataset, showcasing its exceptional accuracy and efficiency in the experimental results.