This project contains a neural network that aims to classify four types of art: painting, iconography, sculpture, and drawing. Instead of a traditional CNN, this is a residual convolutional neural network.
The root folder of this project contains five files + readme. It also contains a folder which contains the training data.
- cleanbadimages.py was used to delete images from the dataset which caused errors during the training. You do not need to run it.
- index.jpg is an image that you can use to test this neural network.
- my_model.zip contains a trained network.
- predict.py can make a prediction on a new image. Just type following to your command line: python predict.py index.jpg. You need to unzip my_model.zip before using predict.py.
- train.py can be used to train a new network on the data.
This model seems to be quite good at classifying sculptures and icons. However, when you give it a painting or a drawing it performs poorly. Committing more data and improving the network architecture is encouraged.