Clothing Attribute Recognition with Semi-supervised Learning

Abstract

Clothing attribute recognition is a challenging task in the field of computer vision and multimedia. In this paper, we propose a semi-supervised method for clothing attribute prediction, which can utilize unsupervised and supervised data together. There are two parts in the proposed model, i.e., the supervised part for training clothing attribute recognition and the unsupervised part for learning the clues of the images themselves. Specifically, we introduce image transformation, i.e., projective transform, as the unsupervised part, and the MSE loss is used to regress the parameters of the transform coefficients. To explore the effectiveness of the proposed semi-supervised method, we design different scales of the unsupervised data to verify it. And the experimental results show the semi-supervised data can obtain good performance and alleviate human labor.