106-facial-landmark-detection

This is just the first version of the code, the latest code has not been organized. The code will be updated soon.

Installation

python 2.7
pytorch 0.4.1
opencv-python
dlib python

Usage

data preprocess

The marked 106-point data should be stored in the .txt format. According to the order of 106 points, the x coordinate of a certain point is stored first, and then the y coordinate is stored. One of the maps corresponds to a txt file with the same name.

python gen_img.py

It requires a origin image storage path imgpath, a txt file storage path landmark, a new txt file storage save_landmark_anno, and an image storage path after the data augmentation landmark_imgs_save_dir.

train

Assign save_landmark_anno to label_path in train.py.

python train.py

The model file is stored as my_net.pkl in the current directory.

test

Assign the test image file path to img_path in test.py.

python test.py