Raftnet: Extract Task-aware Features for Pedestrian Attribute Recognition

Wuhao Zhang, Lizhuang Ma, Yanhao Ge

Abstract: Pedestrian attribute recognition, a multi-task problem, is a popular task in computer vision. Generally, deep learning end-to-end networks to predict attributes are the basic method to solve this problem. To fully use deep neural network, this paper proposes a novel network structure called Raft Block. Raft Block is designed not only to extract task-specific features, but also to share the features of different tasks. Using the Raft Block, we build an end-to-end network Raftnet for pedestrian attribute recognition. We implement experiments on three public datasets, the results prove that the design idea of Raft Block is valid and effective. Specifically, we achieve state-of-art results as 85.64% and 82.79% mean accuracy on Market-1501 and DukeMTMC datasets, and competitive result as 72.53% mAP on PA-100K dataset.