A Convolutional Neural Network with some utilities, such as saving and retraining, for image recognition.
CIFAR-10 contains 60000 images, which are distributed into 10 categories of animals and vechiles.
Using Keras, Tensorflow, Numpy and Matplotlib libraries. Achieving 85% accuracy after 35 epochs.
Run main.py to execute the program, or open Cifar10_preview_example.ipynb in Jupyter Notebook to preview predictions. If you are either retraining or previewing, make sure to have the files "cifar10_model.h5" and "cifar10_model.json" in the execution folder.
Learning Multiple Layers of Features from Tiny Images, Alex Krizhevsky, 2009.