 For me, it was always hard to tell the difference between two painters, Monet and Manet. I'm not an art historian, but I'm a data scientist, and I decided to solve the problem. Well, with orange. Let us load 27 paintings, 15 painted by Monet and 12 by Manet. You can find the link to the data in the description. You can see how import images recognized the two subfolders we have created. If we observe our data in a data table, we see that the names of the folders became our class variable. But as we know from our previous videos, we cannot work with images alone. We need to transform them into their numerical representation. We will use image embedding and stay with the inception version 3 deep network embedder. This embedder was trained on a large collection of images from real life and is able to recognize the motif in the images. This means we can find out whether the two painters painted similar or different motifs. We will measure the similarity between feature profiles of our paintings with distances. Remember to set the distance measure to cosine. Cosine distance will only consider the relative presence of the feature and not its absolute value. Finally, send the computed distance matrix to hierarchical clustering. Wow, the clustering independently discovered two groups that correspond to the two painters. Amazing. But wait a minute, it did make one mistake. There is a Monet's painting that seems to be more similar to Monet's masterpieces than to his own. Let us inspect it. Select the suspicious cluster and send it to image viewer. Set the label to image name to name the paintings with their proper titles. Looks like clustering didn't make a mistake after all. Camille Monet on a bench is indeed very similar to Argentin and in the conservatory. They all portray a man and a lady with a hat sitting on a bench. One thing is very important here. Computation of distances and hence clustering didn't consider class labels. Yet clustering found two clusters of similar images that correspond nicely to our predefined labels. The outlier is there for a reason and it is always worth inspecting why it is there and what are the other similar images.