Multi GAP Neural Network for Emotion Prediction

idea from the paper

Network Description

Used the category classification proposed by Machajdik and Hanbury and the model was trained on dataset proposed by You et al.(2016).

The final categories is divided into 8 categories, and contained amusement, anger, awe, contentment, disgust, excitement, fear, and sadness.

It is usually considered that amusement, awe, contentment and excitement are positive emotion, and the others are negative.

The structure are shown below

Train

1. Download the dataset mentioned above
2. Store images in data folder group by emotion classes.
3. Split data with 'split_data.py'
4. Run object_det.py and places_det.py (Change directory to training_models folder)
5. Run late_fusion2.py (Change directory to training_models folder)

Use Pretrained Models

1. Download FI pretrained weights, Store them in pretrained_models folder
2. Store test images in data/test folder.
3. Run late_fusion2.py

Preview with Streamlit or Gradio

The code for preview on web also available.

Run on streamlit could use the command streamlit run demo_str.py

Run on Gradio could use the command python demo_gra.py