Heterogeneous Face Attribute Estimation: A Deep Multi-Task Learning Approach

Face attribute estimation has many potential applications in videosurveillance, face retrieval, and social media. While a number of methods havebeen proposed for face attribute estimation, most of them did not explicitlyconsider the attribute correlation and heterogeneity (e.g., ordinal vs. nominaland holistic vs. local) during feature representation learning. In this paper,we present a Deep Multi-Task Learning (DMTL) approach to jointly estimatemultiple heterogeneous attributes from a single face image. In DMTL, we tackleattribute correlation and heterogeneity with convolutional neural networks(CNNs) consisting of shared feature learning for all the attributes, andcategory-specific feature learning for heterogeneous attributes. We alsointroduce an unconstrained face database (LFW+), an extension of public-domainLFW, with heterogeneous demographic attributes (age, gender, and race) obtainedvia crowdsourcing. Experimental results on benchmarks with multiple faceattributes (MORPH II, LFW+, CelebA, LFWA, and FotW) show that the proposedapproach has superior performance compared to state of the art. Finally,evaluations on a public-domain face database (LAP) with a single attribute showthat the proposed approach has excellent generalization ability.