A PyTorch implementation of Neural Style Transfer[1].
Download pretrained VGG19 model from the link below and place it in data/.
https://drive.google.com/file/d/1JNrSVZrK4TfC7pFG-r7AOmGvBXF2VFOt/view?usp=sharing
Run the notebook to get the results with default configuration.
You can also customize the configuration. For example,
content_losses_from = [
("conv4_2", 1.0),
]
style_losses_from = [
("conv1_1", 0.2),
("conv2_1", 0.2),
("conv3_1", 0.2),
("conv4_1", 0.2),
("conv5_1", 0.2),
]
losses, style_losses, content_losses = train(content_losses_from=content_losses_from, style_losses_from=style_losses_from,
opt="adam", num_iter=5000, show_every=500, learning_rate=1e-0, pooling="avg",
alpha=1, beta=1e4, tv_weight=0)
content_losses_from: Layers and their contributing weights for content loss.
style_losses_from: Layers and their contributing weights for style loss.
opt: Which optimizer to use. Support adam and LBFGS.
num_iter: Number of iterations to update the image.
show_every: Show progress for every show_every iterations.
alpha: Weight for content loss.
beta: Weight for style loss.
tv_weight: Weight for total variation (L2) loss.
[1] Gatys, L. A., Ecker, A. S., & Bethge, M. (2016). Image style transfer using convolutional neural networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 2414-2423). Retrieved from https://www.cv-foundation.org/openaccess/content_cvpr_2016/papers/Gatys_Image_Style_Transfer_CVPR_2016_paper.pdf