Week 6# Identification of Artists and Movements from Paintings with Machine Learning
This week, we will present the results of LeNet-5 model.
Yann Lecun created LeNet, one of the first convolutional neural networks, which drew attention to deep learning and helped to advance the field. This neural network, which was created in the early 1990s, has primarily been used for handwriting recognition. There are 7 levels in LeNet-5: 3 convolutional layers, 2 subsampling layers, and 2 fully connected layers make up the layer composition.
For our situation, we have 12 classes but LeNet-5 originally designed to recognize 10 classes. For this reason, we recreated the model as follows:
Results
We got the best results when the batch size is 32, epoch is 50 and activation function is relu which is 70.22% for artist identification and 76.93% for movement identification.
If we give an example where we misclassified artists, Claude Monet has 99 misclassified images and about 32% of them are labeled as Pierre Auguste Renoir which is also belongs to Impressionism Art Movement.
If we give an example where we misclassified movements, Neoclassicism has 101 misclassified images and about 41% of them are labeled as Symbolism. See the paintings below:
The right painting originally belongs to Neoclassicism movement, however our model classified as Symbolism movement. These two paintings has very similar color distribution, that might cause misclassification.
