Implementation of the Style-transfer model based on "Perceptual Losses for Real-Time Style Transfer and Super-Resolution"
The stochastic gradient descent function for minimizing the difference between the output image and target image, making the output image (from the transform net) more like the target image
Feature Reconstruction Loss function, penalizes the difference in content
Covariance, capturing the information about which features tend to activate together
Style Reconstruction Loss function, penalizes the difference in style
Pixel Loss, normalized Euclidean distance between the output image and the target
(However, Feature Reconstruction Loss function does a better job at reconstructing fine details, leading to pleasing visual results)
Spatial smoothness
python style.py —-checkpoint-dir ./model/ —-style ./style/wave.jpg
python evaluate.py —-checkpoint ./model/[YOUR_CKPT_NAME].ckpt —-in-path ./examples/content/chicago.jpg —-out-path ./