A TensorFlow project done in Google Colab and Kaggle to classify images of facial emotions. Implemented techniques such as data preprocessing, building and training deep learning models, validating models and plotting performance. The model uses a multi-layer 2D convoluted neural network with batch normalization and max pooling, and uses softmax activation for categorical classification. Performance enhancements are also included, such as reducing learning rate when a plateau in validation loss is detected. The model achieved an accuracy of 62.85%, whereas the highest accuracy published is 76.82%.
View the notebook on Kaggle: https://www.kaggle.com/code/wowthecoder/fer2013/notebook
Kaggle Dataset used: https://www.kaggle.com/datasets/msambare/fer2013
How to use Kaggle Dataset in Google Colab: https://medium.com/unpackai/how-to-use-kaggle-datasets-in-google-colab-f9b2e4b5767c
Note: The validation and test accuracy varies slightly for each run, the differences are about ±1.0%