GoDP: Globally Optimized Dual Pathway deep network architecture for facial landmark localization in-the-wild

Abstract

Facial landmark localization is a fundamental module for pose-invariant face recognition. The most common approach for facial landmark detection is cascaded regression, which is composed of two steps: feature extraction and facial shape regression. Recent methods employ deep convolutional networks to extract robust features for each step, while the whole system could be regarded as a deep cascaded regression architecture. In this work, instead of employing a deep regression network, a Globally Optimized Dual-Pathway (GoDP) deep architecture is proposed to identify the target pixels through solving a cascaded pixel labeling problem without resorting to high-level inference models or complex stacked architecture. The proposed end-to-end system relies on distance-aware softmax functions and dual-pathway proposal-refinement architecture. Results show that it outperforms the state-of-the-art cascaded regression-based methods on multiple in-the-wild face alignment databases. The model achieves 1.84 normalized mean error (NME) on the AFLW database [1], which outperforms 3DDFA [2] by 61.8%. Experiments on face identification demonstrate that GoDP, coupled with DPM-headhunter [3], is able to improve rank-1 identification rate by 44.2% compare to Dlib [4] toolbox on a challenging database. Recent works in human pose estimation [14, 15, 16] employ 2D score maps as the targets for inference. This modification enables gradients back-propagation between stages, allows 2D feedback loops, and hence delivers an end-to-end model. In this new family of methods for key-points localization, the works of Peng [16] et al. and Xiao [32] et al. rely on a DeconvNet [31] architecture to localize facial landmarks.Even though they obtain impressive results by integrating the estimation with recurrent neural networks, the quality of face alignment is intrinsically limited by the lowquality confidence map generated by the DeconvNet as shown in Fig. 1. Wei et al. [15] proposed the convolutional pose machine (CPM), which employs a stacked cascaded architecture to refine body key-point predictions. This cascaded structure has multi-

Wing Loss for Robust Facial Landmark Localisation with Convolutional Neural Networks

“…The value of a pixel in a heat map indicates the probability that its location is the predicted position of the corresponding landmark. To reduce false alarms of a generated 2D sparse heat map, Wu et al propose a distance-aware softmax function that facilitates the training of their dual-path network [63].…”

Adaptive Wing Loss for Robust Face Alignment via Heatmap Regression

“…Loss functions for heatmap regression were rarely studied in previous work. GoDP [65] used a distance-aware softmax loss to assign large penalty on incorrectly classified positive samples, while gradually reducing penalty on missclassified negative samples as the distance from nearby positive samples decrease. The Wing loss [18] is a modified log loss for direct regression of landmark coordinates.…”

Rectified Wing Loss for Efficient and Robust Facial Landmark Localisation with Convolutional Neural Networks

“…fully convolutional network and the hourglass network (Newell et al 2016;Yang et al 2017a;Deng et al 2019b;Bulat and Tzimiropoulos 2017a, b;. To reduce false alarms of a generated 2D sparse heatmap, Wu et al (2018b) proposed a distance-aware softmax function that facilitates the training of their dual-path network. proposed to create a boundary heatmap mask using hourglass network for feature map fusion and showed its beneficial impact on the landmark localisation accuracy.…”

“…We train the plain CNN networks on AFLW using three different loss functions. In addition, we compare the results obtained by these CNN networks with five state-of-the-art baseline algorithms (Feng et al 2017b;Lv et al 2017;Dong et al 2018b, a;Wu et al 2018b). The first baseline method is a multi-view cascaded shape regression model, namely Dynamic Attention Controlled Cascaded Shape Table 2.…”