Dual-Agent GANs for Photorealistic and Identity Preserving Profile Face Synthesis

Jian Zhao, Lin Xiong, Jayashree Karlekar, Jianshu Li, Fang Zhao, Zhecan Wang, Sugiri Pranata, Shengmei Shen, Jiashi Feng
Conference on Neural Information Processing Systems (NIPS) 2017, 2017.12

Synthesizing realistic profile faces is promising for more efficiently training deep pose-invariant models for large-scale unconstrained face recognition, by populating samples with extreme poses and avoiding tedious annotations. However, learning from synthetic faces may not achieve the desired performance due to the discrepancy between distributions of the synthetic and real face images. To narrow this gap, we propose a Dual-Agent Generative Adversarial Network (DA-GAN) model, which can improve the realism of a face simulator’s output using unlabeled real faces, while preserving the identity information during the realism refinement. The dual agents are specifically designed for distinguishing real v.s. fake and identities simultaneously. In particular, we employ an off-the-shelf 3D face model as a simulator to generate profile face images with varying poses. DA-GAN leverages a fully convolutional network as the generator to generate high-resolution images and an auto-encoder as the discriminator with the dual agents. Besides the novel architecture, we make several key modifications to the standard GAN to preserve pose and texture, preserve identity and stabilize training process: (i) a pose perception loss; (ii) an identity perception loss; (iii) an adversarial loss with a boundary equilibrium regularization term. Experimental results show that DA-GAN not only presents compelling perceptual results but also significantly outperforms state-of-the-arts on the large-scale and challenging NIST IJB-A unconstrained face recognition benchmark. In addition, the proposed DA-GAN is also promising as a new approach for solving generic transfer learning problems more effectively.

Link: https://papers.nips.cc/paper/6612-dual-agent-gans-for-photorealistic-and-identity-preserving-profile-face-synthesis

Back

Share

Acceptance of two papers by ICCV’17 Workshop on MS-Celeb-1M: Recognizing One Million Celebrities in the Real World

Panasonic R&D Center Singapore, collaborating with the National University of Singapore (NUS), co-authored two papers that have been accepted by ICCV (International Conference on Computer Vision*) 2017 Workshop on MS-Celeb-1M: Recognizing One Million Celebrities in the Real World, which will be held in Venice, Italy, on October 28th, 2017.

Panasonic R&D Center Singapore marking the first full-year deployment of our AI-based Visitor Counting solution with the latest technology update at the Singapore Flyer’s Time Capsule

Panasonic R&D Center Singapore deployed our AI-based Visitor Counting solution at the iconic Singapore Flyer – Asia’s largest giant observation…

Paper accepted at NIPS2017 and another paper at CoR2017

Panasonic R&D Center Singapore, collaborating with Assistant Professor Feng Jiashi of the National University of Singapore (NUS), co-authored a paper1 titled “Dual-Agent GANs for Photorealistic and Identity Preserving Profile Face Synthesis” that has been accepted at the Conference and Workshop on Neural Information Processing Systems (NIPS) 2017

Panasonic Tops the 2018 Nvidia AI City Challenge as First-Place Winner for Anomaly Detection

Panasonic R&D Center Singapore reprises the lead in Face Recognition Technology by notching up a stunning win for Track 2 (Anomaly Detection) in the 2018 Nvidia AI City Challenge .

Company News