update from old[repo]https://github.com/harvitronix/five-video-classification-methods,https://github.com/oarriaga/face_classification *data are selected from kinetics datasets(including action information and facial expression information) *standard training/testing data are prepared in fold hy_bimodal_sentimentasl_analysis/data/,including (1) how to extract sequeence frame from video (2) the ground truth for training set and testing set
- data include eight classes, the detail of datasets can turn to baiducloud https://pan.baidu.com/s/1bYkPp2qozSQA3qJD-t9n4w
##instructions ###to train models for emotion feature extraction
- download fer2013 datasets with seven emotion classes
- cd multimodal_sentiment_recognition/emotion_recognition/src/utils, modified the path of emotion datasets in datasets.py
- cd multimodal_sentiment_recognition/emotion_recognition/src/, run train_emotion_classifier.py and the trained models are stored into specified path
- cd multimodal_sentiment_recognition/hy_bimodel_sentimental_analysis, runing extract_emotion_features.py to extract facial expression features
###to train models for action feature extraction
- download our collected datasets in baiducloud https://pan.baidu.com/s/1bYkPp2qozSQA3qJD-t9n4w
- cd multimodal_sentiment_recognition/hy_bimodel_sentimental_analysis, running extract_action.py to extract action feature
###to extract synchronous feature from traiend emotion model and traiend action model
- cd multimodal_sentiment_recognition/hy_bimodel_sentimental_analysis, running extract_synchronous_feature.py and storing the features into specified path for traing RNN network
####to train multimodel_model for sentiment analysis
- cd multimodal_sentiment_recognition/hy_bimodel_sentimental_analysis, running training.py