Image Stylizer 

Deep Neural Network Experiment


Image Stylizer is a deep neural network experiment that uses the stylistic elements of one image to draw the content of another.

This project is an simplified implementation of A Neural Algorithm of Artistic Style (Gatys et al) using Python and TensorFlow. It uses a pre-trained model for transfer learning, in this case VGG19 (19-layer VGG Network). I replaced the max pooling layers with average pooling layers and discarded all fully connected layers, as the paper suggests, and after defining heuristics for content and style loss it gave me some very convincing results!

All of the images you see on this page are drawn by Image Stylizer; I've included the content and style images along with the final results for reference. Scroll through the gallery below and take a look!

  • Header Image: Photo of me in the style of Nude Descending A Staircase (Duchamp, 1912)
  • Gallery Image 1: Mona Lisa (Da Vinci, 1506) in the style of Starry Night (Van Gogh, 1889)
  • Gallery Image 3: Nighthawks (Hopper, 1942) in the style of Udnie (Picabia, 1913)
  • Gallery Image 3: Photo of me in the style of Self Portrait Between the Clock and the Bed (Grieg, 1943)
  • Gallery Image 4: Photo of me in the style of Small Worlds I (Kandinsky, 1922)
  •  

     

    © Wei Xiong, 2017