Reflective Regression of 2D-3D Face Shape Across Large Pose

Xuhui Jia, Heng Yang, Xiaolong Zhu, Zhanghui Kuang, Yifeng Niu and Kwok-Ping Chan

Abstract

In this paper we present a novel reflective method to estimate 2D-3D face shape across large pose. We include the knowledge that a face is a 3D object into learning pipeline, and formulate face alignment as a 3DMM fitting problem, where the camera projection matrix and 3D shape parameters are learned by a extended cascaded pose regression framework. In order to improve algorithm robustness to difficult pose, we introduce a reflective invariant metric for failure alert. We investigate the relation between reflective variance and face misalignment error, and find there is strong correlation between them, consequently this finding is exploited to provide feedback to our algorithm. For the sample predicted as failure, we restart them with $better$ initialisation based on explicitly head pose estimation, which enhance the possibility of convergence. Extensive experiments on the challenging $AFLW$ and $AFW$ datasets demonstrate that our approach achieves superior performance than the state-of-the-art methods.

Session

Face and Gesture

Files

Extended Abstract (PDF, 412K)
Paper (PDF, 920K)

DOI

10.5244/C.30.135
https://dx.doi.org/10.5244/C.30.135

Citation

Bibtex

        @inproceedings{BMVC2016_135,
        	title={Reflective Regression of 2D-3D Face Shape Across Large Pose},
        	author={Xuhui Jia, Heng Yang, Xiaolong Zhu, Zhanghui Kuang, Yifeng Niu and Kwok-Ping Chan},
        	year={2016},
        	month={September},
        	pages={135.1-135.14},
        	articleno={135},
        	numpages={14},
        	booktitle={Proceedings of the British Machine Vision Conference (BMVC)},
        	publisher={BMVA Press},
        	editor={Richard C. Wilson, Edwin R. Hancock and William A. P. Smith},
        	doi={10.5244/C.30.135},
        	isbn={1-901725-59-6},
        	url={https://dx.doi.org/10.5244/C.30.135}
        }