Train-Predict-Landmarks-by-DAN

Train Predict Landmarks by Deep Alignment Network (DAN)

based on the code from:
https://github.com/zjjMaiMai/Deep-Alignment-Network-A-convolutional-neural-network-for-robust-face-alignment
modified to our requirments.

tested with:

python 3.6 tensorflow 1.12

Paper:

This code is used for the following research. If you found it usefull, please cite the following document:

https://www.nature.com/articles/s41598-020-58103-6

@article{eslami2020automatic, title={Automatic vocal tract landmark localization from midsagittal MRI data}, author={Eslami, Mohammad and Neuschaefer-Rube, Christiane and Serrurier, Antoine}, journal={Scientific Reports}, volume={10}, number={1}, pages={1--13}, year={2020}, publisher={Nature Publishing Group} }

Following repositories are also used for the mentioned paper:

https://github.com/mohaEs/Train-Predict-Landmarks-by-SFD

https://github.com/mohaEs/Train-Predict-Landmarks-by-MCAM

https://github.com/mohaEs/Train-Predict-Landmarks-by-Autoencoder

https://github.com/mohaEs/Train-Predict-Landmarks-by-dlib

https://github.com/mohaEs/Train-Predict-Landmarks-by-flat-net

data preparation

1- create folders named:
"model_dir"
"prepared_data_test"
"prepared_data_train"
"prepared_data_val"
"results"
"results_post"

2- rearange your data such that for each image, you have a pts file contaring the locations of the landmarks.
format of the pts file is same as the 300w face dataset.
put your data of train, val and test of the folders "data_train/train", "data_train/val" and "data_test".

following image shows an example of folder and files arragament after preparation:

bacth file

now if you want you can use the .bat file for windows terminal which do all of the following parts automatically.
Of course, you can write your own linux shell file.

scaling and preparation

our input image sizes is 256x256 which should be changed to the size suitable for network.

python ./DAN_V2/preprocessing_nocrop.py --input_dir ./data_train/train/ --output_dir ./prepared_data_train/ --img_size 112
python ./DAN_V2/preprocessing_nocrop.py --input_dir ./data_train/val/ --output_dir ./prepared_data_val/ --img_size 112
python ./DAN_V2/preprocessing_nocrop.py --input_dir ./data_test --output_dir ./prepared_data_test/ --img_size 112

train

python ./DAN_V2/DAN_V2_modified.py --dan_stage 1 --data_dir=./prepared_data_train/ --data_dir_test=./prepared_data_val/ --train_epochs=15 --num_lmark 68 --epochs_per_eval=1 -mode train --batch_size=40
python ./DAN_V2/DAN_V2_modified.py --dan_stage 2 --data_dir=./prepared_data_train/ --data_dir_test=./prepared_data_val/ --train_epochs=45 --num_lmark 68 --epochs_per_eval=1 -mode train --batch_size=40

predict

the results will be saved in results folder.

python ./DAN_V2/DAN_V2_modified.py -ds 2 --data_dir=./prepared_data_test/ --model_dir=./model_dir --num_lmark 68 -mode predict

postprocess and rescale

the post processed results will be saved in results_post folder.

python ./DAN_V2/postprocessing_nocrop.py --input_dir ./results/ --output_dir ./results_post/ --img_size 256