Need help with Deep-Alignment-Network-A-convolutional-neural-network-for-robust-face-alignment?
Click the “chat” button below for chat support from the developer who created it, or find similar developers for support.

About the developer

zjjMaiMai
218 Stars 73 Forks 38 Commits 26 Opened issues

Description

This is a Tensorflow implementations of paper "Deep Alignment Network: A convolutional neural network for robust face alignment".

Services available

!
?

Need anything else?

Contributors list

# 342,906
Python
convolu...
Tensorf...
5 commits

Deep Alignment Network: A convolutional neural network for robust face alignment

This is a Tensorflow implementations of paper "Deep Alignment Network: A convolutional neural network for robust face alignment". You can see Original implementation here.


System

  • No Windows !

Getting started

  • Tensorflow 1.7.0
  • OpenCV 3.1.0 or newer

Train Model

  • Download Datasets.
  • Put
    images & pts
    in
    SAME
    folder.
  • Write mirror file. There is a 68 landmark mirror file. download
  • Preprocess.
    shell
    python preprocessing.py --input_dir=... --output_dir=... --istrain=True --repeat=10 --img_size=112 --mirror_file=./Mirror68.txt
    
  • Train model.
    shell
    python DAN_V2.py -ds 1 --data_dir=preprocess_output_dir --data_dir_test=...orNone -nlm 68 -te=15 -epe=1 -mode train
    python DAN_V2.py -ds 2 --data_dir=preprocess_output_dir --data_dir_test=...orNone -nlm 68 -te=45 -epe=1 -mode train
    

Eval Acc

  • Download Datasets for test.
  • Put
    images & pts
    in
    SAME
    folder.
  • Preprocess.
    shell
    python preprocessing.py --input_dir=... --output_dir=... --istrain=False --img_size=112
    
  • Eval model Acc.
    shell
    python DAN_V2.py -ds 2 --data_dir=preprocess_output_dir -nlm 68 -mode eval
    

Results on 300W

  • Speed : 4ms per Image on GTX 1080 Ti
  • Err :
    1.34 %
    on 300W common subset(bounding box diagonal normalization).

Pre-trained Model

TODO:You can download pre-trained model here. This model trained on 300W dataset.

We use cookies. If you continue to browse the site, you agree to the use of cookies. For more information on our use of cookies please see our Privacy Policy.