• Acta Optica Sinica
  • Vol. 39, Issue 11, 1115003 (2019)
Weichi Zhao1, Qijie Zhao1、2、*, Junye Jiang1, and Jianxia Lu1
Author Affiliations
  • 1School of Mechatronic Engineering and Automation, Shanghai University, Shanghai 200444, China
  • 2Shanghai Key Laboratory of Intelligent Manufacturing and Robotics, Shanghai 200444, China
  • show less
    DOI: 10.3788/AOS201939.1115003 Cite this Article Set citation alerts
    Weichi Zhao, Qijie Zhao, Junye Jiang, Jianxia Lu. New Method for Face Landmark Detection Based on Stacked-Hourglass Network[J]. Acta Optica Sinica, 2019, 39(11): 1115003 Copy Citation Text show less
    Structure of stacked-hourglass network
    Fig. 1. Structure of stacked-hourglass network
    Overall framework of face landmark detection method
    Fig. 2. Overall framework of face landmark detection method
    Structural diagram of depth network model
    Fig. 3. Structural diagram of depth network model
    Comparative experiment of face landmark detection on images with large posture changes and face partial occlusion
    Fig. 4. Comparative experiment of face landmark detection on images with large posture changes and face partial occlusion
    CED of different methods for 300W competition dataset with inter-ocular normalization
    Fig. 5. CED of different methods for 300W competition dataset with inter-ocular normalization
    Face feature line heatmaps of 300W competition test set
    Fig. 6. Face feature line heatmaps of 300W competition test set
    Detection results on 300W competition dataset
    Fig. 7. Detection results on 300W competition dataset
    CED for the Menpo competition dataset with face diagonal normalization
    Fig. 8. CED for the Menpo competition dataset with face diagonal normalization
    ConditionMethodCommon subsetChallenging subsetFull set
    Inter-pupil normalizationMethod in Ref. [18]6.6519.799.22
    Method in Ref. [19]5.5016.787.69
    Method in Ref. [20]5.2817.007.58
    Method in Ref. [4]5.6015.407.52
    Method in Ref. [21]5.2513.626.40
    Method in Ref. [22]4.9511.986.32
    Method in Ref. [23]4.5113.806.31
    Method in Ref. [24]4.739.985.76
    Method in Ref. [25]4.808.605.54
    Method in Ref. [26]4.128.354.94
    Method in Ref. [27]3.677.624.44
    FDL-PHR3.227.924.14
    Inter-ocular normalizationMethod in Ref. [27]3.677.624.44
    Method in Ref. [6]3.336.994.05
    Method in Ref. [28]3.346.603.98
    Method in Ref. [29]3.346.563.97
    FDL-PHR3.115.713.62
    Table 1. Error of face landmark detection methods on the 300W test set%
    ConditionMethodN0.08Failure /%
    Inter-ocular normalizationMethod in Ref. [4]0.429410.89
    Method in Ref. [20]0.431210.45
    Method in Ref. [24]0.49875.08
    Method in Ref. [6]0.52124.21
    FDL-PHR0.68932.35
    Table 2. N0.08 and failure rate of face landmark detection methods on the 300W full test set by inter-ocular normalization
    ConditionMethodError /%
    Inter-ocular normalizationMethod in Ref. [21]Method in Ref. [6]FDL-PHR13.8910.028.32
    Table 3. Error of face landmark detection methods on face images with large posture changes and face partial occlusion by inter-ocular normalization
    ConditionMethodN0.08Failure /%
    Inter-ocular normalizationMethod in Ref. [30]0.195538.83
    Method in Ref. [20]0.323517.00
    Method in Ref. [31]0.328113.00
    Method in Ref. [32]0.349712.67
    Method in Ref. [24]0.398112.30
    Method in Ref. [6]0.45326.80
    FDL-PHR0.58052.86
    Table 4. N0.08and failure rate of facial landmark detection methods on the 300W competition dataset by inter-ocular normalization
    ConditionMethodMeanerrorStandard deviationMax error
    Face diagonal normalizationMethod in Ref. [33]0.02050.03400.9467
    Method in Ref. [34]0.01820.01790.4661
    Method in Ref. [35]0.01650.02350.9612
    Method in Ref. [36]0.01590.02010.6717
    Method in Ref. [37]0.02000.07560.7290
    Method in Ref. [38]0.01350.00950.5098
    Method in Ref. [29]0.01380.01570.6312
    Method in Ref. [39]0.01390.02600.9624
    Method in Ref. [9]0.01200.00600.1453
    FDL-PHR0.01990.00710.07184
    Table 5. Error analysis of face landmark detection methods on the Menpo competition dataset by face diagonal normalization
    Weichi Zhao, Qijie Zhao, Junye Jiang, Jianxia Lu. New Method for Face Landmark Detection Based on Stacked-Hourglass Network[J]. Acta Optica Sinica, 2019, 39(11): 1115003
    Download Citation