Identifying Artwork

Inspiration

I really love going to art museums and see if I could guess which works were made by famous artists I had learned about in school. So, I thought it would be a fun to make a program that could do it for me.

What it does

My AI look at artwork from 10 famous artists and classifies the pieces by artist.

How I built it

I used the Keras deep learning library to build a CNN for my specific problem. I Used the pre-built VGG model as a base and added my own neural network layers and adjusted specific parameters to fine-tune my model.

Challenges I ran into

I had very unbalanced data. For some artists I had only 250 pieces and for others I had more than 800, so my model learned how to identify certain artists much more accurately than others. Luckily, I diminished this unbalanced by using class weights.

Accomplishments that I'm proud of

I am glad I final got my accuracy up. The problem of classifying artists is pretty difficult even for humans, so I am happy my model can do it fairly accurately.

What I learned

I learned a lot about neural networks, more specifically CNNs. I also learned a lot about all of Keras's cool features and tools.

What's next for Identifying Artwork

I'd love improve my model by using ImageNet pre-trained weights and experimenting with over and under sampling and maybe expanding it to identify more artists. It also would be fun to use my model in a game where a user can compete against the model.

Built With

• keras
• python