Neural Style Transfer

Artists and painters use their unique skill to create a visual representation by composing a complicated interplay between the content and style of an image. We can use deep neural networks to model and find out the Artist’s algorithmic basis of their process. 

We use Convolutional Neural Networks (CNN) as the underline deep learning neural networks for this task. CNN comprises of layers of small units with visual information that are feeding forward to each layer of the deep learning networks.  The units in each layer is a set of image filters that extract certain features from the input image. As a result, the output of a given layer is a collection of feature maps. 

Since the higher layers in the network represent the high-level content of the input image, the features we capture in higher layers of the networks are regarded as the content representation. The style of an input image is a feature space that is designed to capture texture information. This feature space is the correlations between the different filter response over the spatial extent of the features maps. 

  •  Application of TensorFlow on Neural Style Transfer for Art Generation
Content Image
Style Image
Generated Image

TraNSFER River Drawing

  • Application of TensorFlow on  Neural Style Transfer for Art Generation
Content Image
Style Image
Generated Image

TraNSFER NIGHT Drawing

  • Application of TensorFlow on  Neural Style Transfer for Art Generation
Content Image
Style Image
Generated Image

TraNSFER BirDS Drawing

  • Application of TensorFlow on  Neural Style Transfer for Art Generation
Content Image
Style Image
Generated Image