Deep-Alignment-Network-tensorflow

A re-implementation of Deep-Alignment-Network using TensorFlow

Requirements

Tensorflow 1.5.0 (My version)

Python 3.6

Other commonly used libs for image processing

Dataset formatted like 300-W and Menpo

Usage

Data preparation

cd DAN_TF

mkdir Model

python training\testSetPreparation.py

Training or testing

python train\testDAN.py

Remember to set the STAGE variable and modify the data path in trainDAN.py

MobileNet training or testing

mkdir Model_mobilenet

python trainDAN_mobilenet.py

Remember to set the STAGE variable and modify the data path in trainDAN_mobilenet.py

What's new?

MobileNet: Added on 2018-02-08

References

Original implementation: https://github.com/MarekKowalski/DeepAlignmentNetwork

Another tensorflow implementation: https://github.com/zjjMaiMai/Deep-Alignment-Network-A-convolutional-neural-network-for-robust-face-alignment

Remarks

Note that I have no GPU in this period, so I'm not sure whether this implementation can achieve the same performance as the paper described. However, the functionality of the self defined layers has been tested and I found no problem. If you're interested, you can try my code and train a model. Feel free to raise an issue if you have any question.