Publications of Face frontalization(some with code)

We are only interested in the publications with code

  • Effective Face Frontalization in Unconstrained Images
    • Code: Yes
    • Insight
      • Using only one 3D shape instead of one for each query image
      • Empirically prove that one 3D shape is efficient and doesn’t cause a drop of face recognition performance
    • Drawbacks
      • The rigid/non-rigid ambiguity between head rigid motion and non-rigid facial expressions. Thus the resulting frontalized face might suffer from a neutralization.
      • Their implementation is not stable when the input image is of high resolution.
    • Evaluation scheme
      • Face verification on LFW benchmark
        • They don’t use the SOTA method while testing. SOTA somehow masks the contribution of frontalization.
      • Adience benchmark for gender estimation
  • Large Pose 3D Face Reconstruction from a Single Image via Direct Volumetric CNN Regression
    • Code: Yes
    • insight
      • CNN to learn the mapping from 2D image pixels to 3D facial structure geometry
      • works with one single image
      • Direct regression using CNN. 3DMM is not used.
      • Incorporate landmarks localization to improve the quality
    • Evaluation scheme
      • Cross-datasets experiments: AFLW2000-3D, BU-4DFE, and Florence
      • Normalized mean error to measure the quality of facial mesh:
        • \mathrm{NME}=\frac{1}{N} \sum_{k=1}^{N} \frac{\left|x_{k}-y_{k}\right|_{2}}{d}
  • Towards Large-Pose Face Frontalization in the Wild
    • Code: No
    • Insight
      • deep 3D Morphable Model (3DMM) conditioned Face Frontalization Generative Adversarial Network (GAN), prior 3DMM makes the algo converge faster
      • a new masked symmetry loss to retain visual quality under occlusions
    • Evaluation scheme
      • Qualitative evaluation of frontalization
      • Quantitative evaluation of face recognition


Papers with code for frontalization: