Class Activation and Attention Models for Face Classification
Description of contents
data_preprocessing/: contains the Python scripts needed to tag the images by class (smiling/not smiling, male/female) and upload both the images and the metadata to the AWS S3 bucket.
CNN Research Notebook.ipynb: Jupyter Notebook that reflects setting up basic vanilla CNN image classification architectures.
Image Tagging and Processing.ipynb: Jupyter Notebook that was used to develop the scripts for tagging each image in the FEI dataset by their gender.
MNIST Class Activation Heatmap Example.ipynb: contains the implementation for Class Activation Heatmaps on the MNIST dataset.
vanilla_cnn_faces.py: contains an implementation of VGG16 CNN architecture that achieves strong performance on gender classification of FEI dataset. This was run on a remote GPU-optimized EC2 instance.
Class Activation Map Faces.ipynb: contains VGG16 CAM implementation on the FEI dataset, along with some results and heatmap examples.
Vanilla Self Attention.ipynb: contains VGG16 Multi-Head Augmented Attention model implementation on the FEI dataset. Note that I have annotated which portions of the code are borrowed from another Github repository.