rbbernardino / face-recognition-eval

Several face recognition algorithms evaluation for research purpose.

face-recognition-eval

Several face recognition algorithms evaluation for research purpose.

tested models

pyImageSearch code

• OpenCV + dlib

pack2: Seetaface

About

- China National Computer vision group
- CNN (convolutional neural net)
- VIPLFaceNet (network)
  - 7 convolutional layers
  - 2 fully-connected layers
  - input size of 256x256x3
- tailored from AlexNet to be fast and accurate
  - 40% more accurate
  - 80% faster training
  - 40% faster feature extraction 
- based on cutting edge findings in deep learning research
- 2016 last updated
- pre-trained model
  - 1.4M face images
  - 16K subjects (Mongolians and Caucasians)
- 2048-dimensional feature vector
- on i7-3770 CPU: feature extraction in 115ms

How to Run

1. Detection
   • go

pack3: SeetaFace2

About

a.........

How to Run

b...

Evaluation Protocol

Recognition

Common Configs

• 4x people
• 10x images from each people (several positions)