ComboLoss for Facial Attractiveness Analysis with Squeeze-and-Excitation Networks

• DOI: 10.48550/arxiv.2010.10721
• Publication Date: 2020-01-01
• Authors: Lu Xu; Jinhai Xiang
• Keywords: Code (set theory); Facial attractiveness; Representation; Feature (linguistics)

Loss function is crucial for model training and feature representation learning, conventional models usually regard facial attractiveness recognition task as a regression problem, adopt MSE loss or Huber variant supervision to train deep convolutional neural network (CNN) predict score. Little work has been done systematically compare the performance of diverse functions. In this paper, we firstly analyze under Then novel named ComboLoss proposed guide SEResNeXt50 network. The method achieves state-of-the-art on SCUT-FBP, HotOrNot SCUT-FBP5500 datasets with an improvement 1.13%, 2.1% 0.57% compared prior arts, respectively. Code are available at https://github.com/lucasxlu/ComboLoss.git.