Stage-wise Face Alignment using Global and Local Regressors

This is a caffe-python implementation on Windows 10 for face alignment.

We implemented two-kind of methods.

Method1 repeat global and local regression after initialization regression

Method2 repeat local refinement regression after initialization regression

Evaluation on 300w public test set

Usage

For Training

1. Clone the repository

git clone https://github.com/hyunsungP/facelignmentregression

2. make data files (.h5)

make_wild_input.py

and so on.

3. make data file list
   Refer to models/list_train_*.txt

4. training
   On console window with caffe

caffe train --solver=models/ZF_solver.prototxt --gpu=0

Other network are same.

For Testing

Change prototxt path in the source code.

test_300w_public.py

Other models will be uploaded.