Recognize the Silhouette Attributes of T-shirts based on Mask R-CNN

Clothing is currently one of the most commercially profitable industries in the world, and clothing fashion analysis mines business intelligence information on massive data for the fashion industry, which has received tremendous attentions. As the basic task of fashion analysis, clothing silhouette attribute recognition is of great significance, which determines the accuracy of subsequent analysis. However, the clothing engineering knowledge system is highly professional, and the silhouette attributes are different for different clothing types. What’s more, clothing distortion and occlusion often appear in the fashion images, which brings challenges to the recognition. In this paper, we summarize the key silhouette attributes of T-shirts, and recognize the attributes by instance segmentation based on Mask R-CNN. Through the experiments over the online clothing shopping dataset, we demonstrate the effectiveness of our proposed solution.