Facial landmark detection aims at locating a sparse set of fiducial facial key-points. Two significant issues (i.e., Intra-Dataset Variation and Inter-Dataset Variation) remain in datasets which dramatically lead to performance degradation. Specifically, dataset variations will lead to severe over-fitting easily and perform poor generalization in recent in-the-wild datasets which severely harm the robustness of facial landmark detection algorithm. In this study, we show that model robustness can be significantly improved by lever-aging rich variations within and between different datasets. This is non-trivial because of the serious data bias within one certain dataset and inconsistent landmark definitions between different datasets, which make it an extraordinarily tough task. To address the mentioned problems, we proposed a novel Deep Coupling Neural Network (DCNN), which consists of two strong coupling sub-networks, e.g., Dataset-Across Network (DA-Net) and Candidate-Decision Network (CD-Net). In particular, DA-Net takes advantage of different characteristics and distributions across different datasets, while CD-Net makes a final decision on candidate hypotheses given by DA-Net to leverage variations within one certain dataset. Extensive evaluations show that our approach dramatically outperforms state-of-the-art methods on the challenging 300-W and WFLW dataset (C) 2019 Elsevier Ltd. All rights reserved.
