cdvel / face-recognition-notes

Notes on face detection, verification and recognition

• FaceNet

  • FaceNet: A Unified Embedding for Face Recognition and Clustering

    • Use of triplet loss as ConvNet loss function
    • Face embeddings, that map faces as feature vectores. Easy to compare and measure similarity.
    • LFW dataset accuracy = 99.63%. YouTube Faces DB = 95.12%.

  • Triplet loss

    • Triplet Loss - Special applications: Face recognition & Neural style transfer | Coursera
    • Triplet Loss and Online Triplet Mining in TensorFlow

  • OpenFace

    • Python and Torch implementation of face recognition based on FaceNet
    • Implementation of face recognition with deep neural networks

  • Comparing embeddings

    • Comparing embeddings, FaceNet, euclidean distance between embeddings

• Facial landmark detection

  • One Millisecond Face Alignment with an Ensemble of Regression Trees

• Ageing, adding features, beards etc.

  • Deep Feature Interpolation (CVPR 2017) github

• Frontalization

  • Effective Face Frontalization in Unconstrained Images