One of the first things I did when I learned about Neural Nets and image classifiers was make some art. After a bit of Googling I determined Neural Style Transfer was what I was looking to do. A few years back a film came out called Loving Vincent, which was an animated film in which each frame was painted in the style of Van Gough’s work. That is basically what style transfer is, except I am making a computer do it. So I followed a Medium tutorial here: https://medium.com/tensorflow/neural-style-transfer-creating-art-with-deep-learning-using-tf-keras-and-eager-execution-7d541ac31398 to utalize Tensorflow along with a VGG19 CNN, to apply the style transfer. The VGG19 CNN was created and trained on Imagnet, a database of over 10 million images for deep learning, by a team at Oxford known as Visual Geomotry Group. VGG19 is an insanely powerful image classifier. With it I will be able to use Tensorflow to classify parts of the style image, and apply the same techniques the artist uses on similarrly classified parts of the image to be styled (Content Image). I’ll show you some examples!
This first piece was made by tranfering the art style of my friend with her real life headshots! (shoutout to @DanaJoyJay on insta for these images)
Content Image
Style Image
Transformation Gif
Final Product!
The transformation gif is how the computer styles the image over time. This gif is 1000 itterations of learning!
Another fun Image I transformed was one of the Boulder Flatirons. They are an amazing rock formation in the heart of our country, and what better style to impose than that of Dutch Painter we were just talking about! Van Gough!
Content Image
Style Image
Transformation Gif
Final Product
And lastly one of my friends really likes the artist Edward Hopper and she took a picture that reminded her very much of one of his paintings, so I did a style transfer there too!
Content Image
Style Image
Final Product
This was an amazing expirience for me to dip my feet into more machine learning applications and products and I am more than excited to do more!
Link to Github Repo: https://github.com/BobBriksz/TensorFlowArt