VATA: Video Attention, Text Attention

This is the implementation of our final model for the QIA 2020 Challendge on Kaggle.

How to run:

1. Download the dataset from Kaggle.
2. Use MTCNN and OpenCV to crop out the faces from the video.
3. Transform frames into PyTorch tensors.
4. Run the training file: python fe_train.py

Structure