Face Detection using a custom-made MiniVGG Model so that it can be trained for the whole dataset without exhausting gpu resources Improved accuracy using raytune hyperparameter tuning Fixed the issue of unnormalized loss which was blowing up in the start.