https://github.com/justanhduc/brushstroke-parameterized-style-transfer

https://github.com/justanhduc/brushstroke-parameterized-style-transfer

Last synced: about 2 months ago
JSON representation

• Host: GitHub
• URL: https://github.com/justanhduc/brushstroke-parameterized-style-transfer
• Owner: justanhduc
• Created: 2021-04-03T10:17:22.000Z (over 3 years ago)
• Default Branch: master
• Last Pushed: 2021-05-25T07:34:12.000Z (over 3 years ago)
• Last Synced: 2024-06-08T11:36:43.201Z (4 months ago)
• Language: Python
• Size: 4.5 MB
• Stars: 39
• Watchers: 3
• Forks: 3
• Open Issues: 3
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Parametrized-brushstroke Image Style Transfer

This is a Pytorch implementation of the paper 

["Rethinking Style Transfer: From Pixels to Parameterized Brushstrokes"](http://arxiv.org/abs/2103.17185).



  

  

  



## Prerequisites

[Pytorch](https://pytorch.org/) (>= 1.8 (required))

[Pytorch cluster](https://github.com/rusty1s/pytorch_cluster)

[Neural monitor](https://github.com/justanhduc/neural-monitor) (For logging)

## Running the code

```

python main.py /path/to/content/image /path/to/style/image

```

Type `python main.py` to see all available options.

## Similarities/differences to the official implementation

Similarities

- The renderer and the initialization schemes are no-brainer adaptations from the 

official repo.

- The training schemes (including learning rates and optimizers) are the same.

- The visual results look largely similar.

Differences

- Some minor details regarding the style losses are different.

## References

[Official Tensorflow implementation](https://github.com/CompVis/brushstroke-parameterized-style-transfer)

[VGG pretrained weight file](https://github.com/ftokarev/tf-vgg-weights/raw/master/vgg19_weights_normalized.h5)