Image Stylization Using a Convolutional Neural Network

Dependencies

• Python 3.6
• Keras
• TensorFlow
• NumPy

Usage

python3 style-transfer.py <content_image> <style_image> <output_filename> <optional_parameters>

An example of this is:

python3 style-transfer.py images/content/stata.jpg images/style/starry_night.jpg images/generated/starry_night_stata

The entire run takes about 30 mins to 1 hour depending on the image, the weights of the content and style loss, and the style layers used in the VGG19 CNN.

Optional Parameters

We have provided the option to run the Python script with different parameters, such as:

--iterations <number of iterations>: this is the maximum number of iterations that the gradient descent will go through before the script ends

--style_weight <style weight value>: this is the weight associated with the style loss as part of the total loss function

--content_weight <content weight value>: this is the weight associated with the content loss as part of the total loss function