Overview / Usage

Neural Style Transfer can be used to add cool design to your own images by choosing from any style image available.

Methodology / Approach

Neural Style Transfer exploits the learning property of neural networks that is the initial layers of neural network identify the edges and gradually the final layers identify a small component of an image.For example in face recognition initial layers identify edges while final layers identify eyes, nose etc.
In this approach we start with a initial random noisy image and gradually train it to become more like the content and the style image using the triplet loss function.
A unique point here is that we do not train the neural network but rather we alter the pixel values of the initial random image to minimise the loss function.

Technologies Used

Python
TensorFlow
Ipython