Provides installation guide.

Details on structuring and preparing the data.

Details on how to run the code.

For most users, pip3 install torch torchvision should work. If you are using Anaconda, conda install pytorch torchvision -c pytorch should work.

Create a virtual environment using Conda or Virtualenv and install all the dependencies using requirement.txt file.

The first step is to create the folder heirarchy. The dataset is originally split by subject. You should create separate directories per class.

Biovid

• 0 # class folder (BL1)
  • 071309_w_21-BL1-082 # Subject folder
    • img_00001.jpg #face images
    • .
    • img_00075.jpg #face images
  • .
  • 110810_m_62-BL1-094
    • img_00001.jpg #face images
    • .
    • img_00075.jpg #face images*
• 4 # Class folder (PA4)
  • 071309_w_21-PA4-006 # subject folder
    • img_00001.jpg #face images
    • .
    • img_00075.jpg #face images
  • 071614_m_20-PA4-010
    • img_00001.jpg #face images
    • .
    • img_00075.jpg #face images

Create annotaions files using create_annotationstxt.py file. The annotations should be in the following format. class_folder/subject_folder start_frame end_frame class label e.g. 4/073109_w_28-PA4-062 0 75 1 The create_annotationstxt.py file also provides the code for k-fold annotations.

The 'pkdot_kfold.py' file requires the paths for the pretrained teacher models.