HAVANA: Hierarchical and Variation-Normalized Autoencoder for Person Re-identification
This work aims to improve person re-identification accuracy for video surveillance systems by developing a method that is robust to image variations.
This paper addresses person re-identification by proposing HAVANA, a Hierarchical and Variation-Normalized Autoencoder. HAVANA learns features robust to intra-class variations without extra supervision, and it introduces a Jensen-Shannon triplet loss for contrastive distribution learning.
Person Re-Identification (Re-ID) is of great importance to the many video surveillance systems. Learning discriminative features for Re-ID remains a challenge due to the large variations in the image space, e.g., continuously changing human poses, illuminations and point of views. In this paper, we propose HAVANA, a novel extensible, light-weight HierArchical and VAriation-Normalized Autoencoder that learns features robust to intra-class variations. In contrast to existing generative approaches that prune the variations with heavy extra supervised signals, HAVANA suppresses the intra-class variations with a Variation-Normalized Autoencoder trained with no additional supervision. We also introduce a novel Jensen-Shannon triplet loss for contrastive distribution learning in Re-ID. In addition, we present Hierarchical Variation Distiller, a hierarchical VAE to factorize the latent representation and explicitly model the variations. To the best of our knowledge, HAVANA is the first VAE-based framework for person ReID.