A CLIP based pytorch implementation on facial expression recognition (KMU-FED), achieving an average accuracy of 97.36% in KMU-FED.
This is the official repository for the paper "LEVERAGING VISION LANGUAGE MODELS FOR FACIAL EXPRESSION RECOGNITION IN DRIVING ENVIRONMENT".
- KMU-FED dataset from https://cvpr.kmu.ac.kr/KMU-FED.html
-For KMU-FED dataset: 'python formatdescription.py' to add the text description to the images and save them as CSV format, then, put them in the "data" folder.
'python preprocess_KMUFED.py' to preprocess the image and text data.
Mode 0: Image features only.
Mode 1: Image and text features.
- KMU-FED dataset: python 10fold_train.py
- python KMUconfmtrx.py --mode 1
We use 10-fold Cross validation in the experiment.
- Model: CLIVP-FER ; Average accuracy: 97.364%