革新知能統合研究センター

Title: Image Disentangled Representation and Multi-Attribute Transfer

Speaker: Liqing Zhang
Department of Computer Science and Engineering
Shanghai Jiao Tong University, Shanghai, China

Abstract: Disentangled representation and generative adversarial networks facilitate image-to-image style transfer problem by specifying the attribute vector to the target domain attributes. In this talk, we present a novel disentangled representation formulation by projecting images to latent units, grouped feature channels of Convolutional Neural Network, to disassemble the information between different attributes. Thanks to disentangled representation, we can transfer attributes according to the attribute labels and moreover retain the diversity beyond the labels, namely, the styles inside each image. Several experimental results such as the facial expression transfer and clothing recommendations will be given to the performance and generalizability of the proposed model.

Bio: Liqing Zhang received the Ph.D. degree from Sun Yatsen University, China, in 1988. He is a full professor and Chair of Department of Computer Science and Engineering, Shanghai Jiao Tong University. He serves as the head of SJTU-Versa joint lab for Brain Inspired Computing and AI. His current research interests cover cortical network modelling, computer vision, statistical machine learning.