Coarse-to-Fine Auto-Encoder Networks (CFAN) for Real-Time Face Alignment

Abstract

Accurate face alignment is a vital prerequisite step for most face perception tasks such as face recognition, facial expression analysis and non-realistic face re-rendering. It can be formulated as the nonlinear inference of the facial landmarks from the detected face region. Deep network seems a good choice to model the nonlinearity, but it is nontrivial to apply it directly. In this paper, instead of a straightforward application of deep network, we propose a Coarse-to-Fine Auto-encoder Networks (CFAN) approach, which cascades a few successive Stacked Auto-encoder Networks (SANs). Specifically, the first SAN predicts the landmarks quickly but accurately enough as a preliminary, by taking as input a low-resolution version of the detected face holistically. The following SANs then progressively refine the landmark by taking as input the local features extracted around the current landmarks (output of the previous SAN) with higher and higher resolution. Extensive experiments conducted on three challenging datasets demonstrate that our CFAN outperforms the state-of-the-art methods and performs in real-time(40+fps excluding face detection on a desktop).

Sign-Correlation Partition Based on Global Supervised Descent Method for Face Alignment

“…Ren et al [33] build local binary features by learning regression random forest for each landmark independently, and then learn a global cascaded linear regressor with pre-built binary features. Deep Neural Networks [16,24,25] have also been studied for face landmark detection. DNNs-based methods fuse the feature description and networking training in a unified framework, but it is still a very hard task to tune many free parameters.…”

“…The choice and learning of shape-indexed features are also studied [15][16][17]. A series of regression methods have been employed into cascaded regression framework to deal with over-fitting and local minima problems in the wild condition, including ridge regression [18], Support Vector [19], Gaussian process [20,21], Random Forest voting [14,22,23], Deep Neural Nets [16,24,25], and project-out cascaded regression [26].…”

An Architecture of Deep Learning Method to Predict Traffic Flow in Big Data

“…Zhang and Chen [2] has presented a paradigm of distributed learning for Restricted Boltzmann Machines (RBMs) and also an algorithm of a back propagation (BP) by using Map Reduce and a programming model parallel. The author has gone through the Deep Belief Nets (DBNs) and RBNs.…”

“…Exclusively using the back propagation along with the gradient-descent based optimization is proved to be slow and sometimes halts in deep networks, as noted by James Martens et al [1]. Hence greedy layer wise unsupervised algorithm is considered for training, as suggested by Hinton et al [2], to train the network starting from hidden layer 1. Finally, back propagation is only used to fine-tune the network from the inverse side of shrouded layers and proceed till the principal concealed layer.…”

“…The main objective associated with the prediction of traffic information is to provide the traffic information to each and every vehicle associated with one network in order to optimize the traffic congestion issues in vehicular adhoc networks (VANET) [1] [2]. The recent conventional studies highlights that now a days the deployment of Intelligent Transport Systems (ITSs) has improved the traffic operation efficiency with respect to better travel decision making, reducing carbon emissions optimization of level of congestion.…”