Art-Generation-with-neural-networks

A style transfer python script, based on deep learning, that is able to apply a new style to an image while still preserving its original content.

Implementation Details

This implementation of style transfer uses TensorFlow to train a style transfer network, using the VGG19 pretrained model on ImageNet.

Step1 :

Before trying the model, download the pretrained model from here and extract it to the folder ('pretrained-model').

Step 2 :

Add both your content image and style image to the folder ('images') and make sure to resize them to 400x300 and rename them respectively to my_content and my_style or you can just edit the code to use the names you like :

content_image = scipy.misc.imread("images/your_image_here.jpg")
style_image = scipy.misc.imread("images/your_style_here.jpg")

Step 3 :

in the line 69 of style.py , the number of itereration by default is 400 , if you're using your own datasets, try this number and then more iterations to check the best performance !

def model_nn(sess, input_image, num_iterations = 400):

Step 4 :

Make sure to install all the required libs and then run :

python style.py

Exemple :

the input images (content and style) :

The results : - iteration 0 :

- iteration 400:

you can check in the output folder the images generated after each 20 iteration .

Attributions

• Some readme/docs formatting was borrowed from longstorm fast style transfer