The project was to implement a light weight command line application to create various different CNN architectures. The application is implemented with PyTorch, NumPy and the standard Python library and utilizes the deep learning technique called "transfer-learning", where pre-trained feature extractors are used as the back bone for the application pipe line. The appliation was tested with various different models on the "Flower" dataset (https://s3.amazonaws.com/content.udacity-data.com/nd089/flower_data.tar.gz).