Train Predict Landmarks by Multi-context attention model
based on the code:
https://github.com/wbenbihi/hourglasstensorlfow
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-DAN
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
- generate the text files contains the information of the landmarks for each image in the train folder.
- set the annotaion file path and other configs such as epochs, stack levels and etc. in config file.
Following image shows an example of folders, annotation text file and config file for training. more information is availble at the original repositort mentioned above.
python train_launcher.py
for reading the logs:
tensorboard --logdir=./Logs/ <br> tensorboard --logdir=./Logs-test/
use following shell and asssinging the input directoty, path of the trained model and output directory
python pr_predict.py --input_dir ./temp_test_png --checkpoint ./hg_refined_200 --output_dir ./Results/
images and text files of the prediction are available at the Results folder:
