10th EAI International Conference on Mobile Multimedia Communications

Research Article

Masked Loss Residual Convolutional Neural Network For Facial Keypoint Detection

Download11 downloads
  • @INPROCEEDINGS{10.4108/eai.13-7-2017.2270651,
        author={Junhong Xu and Shaoen Wu and Shangyue Zhu and Hangqing Guo and Honggang Wang and QIng Yang},
        title={Masked Loss Residual Convolutional Neural Network For Facial Keypoint Detection},
        proceedings={10th EAI International Conference on Mobile Multimedia Communications},
        publisher={EAI},
        proceedings_a={MOBIMEDIA},
        year={2017},
        month={12},
        keywords={facial keypoint detection cnn deep learning},
        doi={10.4108/eai.13-7-2017.2270651}
    }
    
  • Junhong Xu
    Shaoen Wu
    Shangyue Zhu
    Hangqing Guo
    Honggang Wang
    QIng Yang
    Year: 2017
    Masked Loss Residual Convolutional Neural Network For Facial Keypoint Detection
    MOBIMEDIA
    EAI
    DOI: 10.4108/eai.13-7-2017.2270651
Junhong Xu1, Shaoen Wu,*, Shangyue Zhu1, Hangqing Guo1, Honggang Wang2, QIng Yang3
  • 1: Ball State University
  • 2: University of Massachusetts Dartmouth
  • 3: Montana State University
*Contact email: swu@bsu.edu

Abstract

This paper presents an e ective and novel CNN-based deep learning solution, named Masked Loss Residual Convolutional Neural Net- work (ML-ResNet), to facial keypoint detection on the datasets that have missing target labels. The core of the ML-ResNet solution is a masked loss objective function that ignores the error in predicting the missing target keypoints in the output layer of a CNN. To com- pensate for the loss induced by the masked loss objective function to prevent over tting, we design a data augmentation strategy in ML-ResNet to increase the number of training data. The perfor- mance of ML-ResNet has been evaluated on the image dataset from Kaggle Facial Keypoints Detection competition, which consists of 7,049 training images, but with only 2,140 images that have full tar- get keypoints labeled. In the experiments, ML-ResNet is compared to a pioneer literature CNN facial keypoint detection work. The experiment results clearly show that the proposed ML-ResNet is robust and advantageous in training CNNs on datasets that have missing target values. ML-ResNet can improve the learning time by 30% during the training and the detection accuracy by eight times in facial keypoint detection.