Week 4 — Image Transformation according to Art Style
Hi! Last week, we have run CycleGAN first. Today, we are going to talk about selective clustering to improve our CycleGAN model.
So far, we have worked on Cycle-GAN. We have tried our model with different artist. The model works properly for most of the artist’s arts. After that, our next aim is to add different solutions to improve the results of our model. We overthink this problem and decided to implement an outlier detection. We will use selective clustering as the outlier detection for our model. This clustering method takes the role to eliminate artist's outlier arts.
Selective classification (or rejection based classification) has been proved useful in many solutions to problems. With the help of deep learning techniques, we will extract content-style features from a pre-trained convolutional network for the paintings. By proposing a rejection mechanism under the Bayesian framework, we will focus on selecting style-oriented representative paintings of an artist, which is an interesting and challenging cultural heritage application. Two kinds of samples are rejected during the rejection based robust continuous clustering process. Representative paintings are selected during the selective clustering phase. Visual qualitative analysis on small painting set and large scale quantitative experiments on a subset. Also, we will research different clustering method if we fail in selective clustering. For example, we can try the oriented clustering method that object-based clustering method to style based method.
Contributors:
