A Structured Feature Learning Model for Clothing Keypoints Localization

Visual fashion analysis has attracted many attentions in the recent years. Especially, as a fundamental technology, clothing keypoints localization has great application potential. However, most of researchers seldomly consider the inherent structural information of clothing and process clothing images as ordinary images. In this paper, a Structured Feature Learning Model (SFLM) is proposed to exploit the structure information of clothing, which models the relationships among clothing keypoints on the feature layer and passes the information among the neighboring keypoints. The model introduces the bi-directional tree and the geometrical transform kernels to construct the information flow and capture the relationships respectively. Therefore, the clothing keypoints features and their relationships can be well jointly learned. The proposed model improves feature learning substantially. We demonstrate that our proposed model has an excellent ability to learn advanced deep feature representations for clothing keypoints localization. Experimental results show that the proposed model outperforms the state-of-the-arts on the DeepFashion and FLD dataset. The code will be available at https://github.com/suyuyiS/SFLM.