awentzonline/image-analogies
A neural network implementation that generates new images by learning analogies from example image pairs using VGG16 feature maps.

This repository implements the Image Analogies paper using deep learning. It extracts feature maps from VGG16 (a convolutional neural network) and uses PatchMatch for efficient approximate nearest-neighbor patch matching between images. The system blends features from a source/target pair to apply analogous transformations to a new image, combining CNN features with Markov Random Field constraints for coherent image synthesis. It runs on TensorFlow or Theano and was adapted from the Keras neural style transfer example.