Week #3 — Identification of Artists by Their Paintings with Machine Learning
In this week, after the examined related works about CNN, we create a dataset and tried on the ResNet-152 model. According to our research, ResNet-152 has much better accuracy compared to ResNet-34 and ResNet-50 at related work. Thus we worked on this model and observed the results.
Platform
We decided to use Google Colab as a platform. It provides a common workspace and GPU resource.
Dataset
We have created a dataset of paintings by 5 known painters. These painters are Vincent Van Gogh, Paul Gauguin, Pablo Picasso, Edgar Degas, Albrecht Dürer.This data is taken from Best Artworks of All Time Kaggle dataset. It includes 300 paintings per artist.
Data Augmentation
Images are flipped, rotated, and resized.
Results
After the augmentation part, we ran our model.
According to confusion matrix, it can be seen that the toughest dual is Picasso and Van Gogh. The reason can be about their style is more similar than the others.
When we look at Dührer, our model can predict it better. The reason can be his style is more recognizable for our model because his paintings are black and white.
In our future work, we want to expand our work with different artists by increasing our data set. We will also seek solutions for the mixed artists (Van Gogh, Picasso, and Gauguin) that we observed in the confusion matrix.
Stay Tune!
References
