Skip to content

Latest commit

 

History

History
102 lines (94 loc) · 4.75 KB

README.md

File metadata and controls

102 lines (94 loc) · 4.75 KB
Raw

Data and Data Prepareing Tool

  • BP4D
    • Image should be placed in data/BP4D/img (or you change the dataset path in config/BP4D_config.yaml)
    • For example: data/BP4D/img/F001/T1/2440.jpg
    • Image path, AU label, AU Relation label (for egde feature supervision in our paper) and AU class weight (for the weighted asymmetric loss) lists are in data/BP4D/list/
    • data/BP4D/list/
├── data/BP4D/list/BP4D_test_img_path_fold1.txt
├── data/BP4D/list/BP4D_test_img_path_fold2.txt
├── data/BP4D/list/BP4D_test_img_path_fold3.txt
├── data/BP4D/list/BP4D_test_label_fold1.txt
├── data/BP4D/list/BP4D_test_label_fold2.txt
├── data/BP4D/list/BP4D_test_label_fold3.txt
├── data/BP4D/list/BP4D_train_img_path_fold1.txt
├── data/BP4D/list/BP4D_train_img_path_fold2.txt
├── data/BP4D/list/BP4D_train_img_path_fold3.txt     
├── data/BP4D/list/BP4D_train_label_fold1.txt
├── data/BP4D/list/BP4D_train_label_fold2.txt
├── data/BP4D/list/BP4D_train_label_fold3.txt
├── data/BP4D/list/BP4D_train_AU_relation_fold1.txt
├── data/BP4D/list/BP4D_train_AU_relation_fold2.txt
├── data/BP4D/list/BP4D_train_AU_relation_fold3.txt
├── data/BP4D/list/BP4D_weight_fold1.txt
└── data/BP4D/list/BP4D_weight_fold2.txt
└── data/BP4D/list/BP4D_weight_fold3.txt
  • DISFA
    • Image should be placed in data/DISFA/img (or you change the dataset path in config/DISFA_config.yaml)
    • For example: data/DISFA/img/SN001/0.png
    • Image path, AU label, AU Relation label (for egde feature supervision in our paper) and AU class weight (for the weighted asymmetric loss) lists are in data/DISFA/list/
    • data/BP4D/list/
├── data/DISFA/list/DISFA_test_img_path_fold1.txt
├── data/DISFA/list/DISFA_test_img_path_fold2.txt
├── data/DISFA/list/DISFA_test_img_path_fold3.txt
├── data/DISFA/list/DISFA_test_label_fold1.txt
├── data/DISFA/list/DISFA_test_label_fold2.txt
├── data/DISFA/list/DISFA_test_label_fold3.txt
├── data/DISFA/list/DISFA_train_img_path_fold1.txt
├── data/DISFA/list/DISFA_train_img_path_fold2.txt
├── data/DISFA/list/DISFA_train_img_path_fold3.txt  
├── data/DISFA/list/DISFA_train_label_fold1.txt
├── data/DISFA/list/DISFA_train_label_fold2.txt
├── data/DISFA/list/DISFA_train_label_fold3.txt
├── data/DISFA/list/DISFA_train_AU_relation_fold1.txt
├── data/DISFA/list/DISFA_train_AU_relation_fold2.txt
├── data/DISFA/list/DISFA_train_AU_relation_fold3.txt
├── data/DISFA/list/DISFA_weight_fold1.txt
├── data/DISFA/list/DISFA_weight_fold2.txt
├── data/DISFA/list/DISFA_weight_fold3.txt

Tools for prepareing data

After getting the datasets (BP4D and DISFA) , you can use our data processing tools in tool/.

    tool/
    ├── BP4D_image_label_process.py
    ├── DISFA_image_label_process.py
    ├── BP4D_calculate_AU_class_weights.py
    ├── DISFA_calculate_AU_class_weights.py
    ├── BP4D_deal_AU_relation.py
    ├── DISFA_deal_AU_relation.py

Before testing or training on BP4D, run:

cd tool/
python BP4D_image_label_process.py
python BP4D_calculate_AU_class_weights.py
python BP4D_deal_AU_relation.py

Before testing or training on DISFA, run:

cd tool/
python DISFA_image_label_process.py
python DISFA_calculate_AU_class_weights.py
python DISFA_deal_AU_relation.py

BP4D_image_label_process.py and DISFA_image_label_process.py are two files to get img_path_fold*.txt and label_fold*.txt from source data files. For example,

  • to get BP4D_train/test_img_path_fold*.txt and BP4D_train/test_label_fold*.txt, run:
python BP4D_image_label_process.py

BP4D_calculate_AU_class_weights.py and DISFA_calculate_AU_class_weights.py are two files to get weight_fold*.txt from label_fold*.txt. For example,

  • to get BP4D_weight_fold*.txt, run:
python BP4D_calculate_AU_class_weights.py

BP4D_deal_AU_relation.py and DISFA_deal_AU_relation.py are two files to get train_AU_relation_fold*.txt from label_fold*.txt. For example,

  • to get DISFA_train_AU_relation_fold*.txt, run:
python DISFA_deal_AU_relation.py