https://github.com/kahsolt/image-style-transfer

Images style transfer via many pretrained models by Gatys' method
https://github.com/kahsolt/image-style-transfer

Last synced: 5 months ago
JSON representation

Images style transfer via many pretrained models by Gatys' method

• Host: GitHub
• URL: https://github.com/kahsolt/image-style-transfer
• Owner: Kahsolt
• License: mit
• Created: 2024-10-12T16:01:21.000Z (over 1 year ago)
• Default Branch: master
• Last Pushed: 2025-01-25T09:59:59.000Z (12 months ago)
• Last Synced: 2025-01-30T14:19:51.026Z (12 months ago)
• Language: Python
• Size: 1.02 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# image-style-transfer

    Images style transfer via many pretrained models by Gatys' method

----

ℹ Extending Leon A. Gatys' paper ["Image Style Transfer Using Convolutional Neural Network" (CVPR 2016)](https://www.cv-foundation.org/openaccess/content_cvpr_2016/papers/Gatys_Image_Style_Transfer_CVPR_2016_paper.pdf) to other pretrained models far beyond vgg19 ;)

### Experiments

> All tests are run on a single RTX 3060.

| content image | style image |

| :-: | :-: |

| ![content](./img/Tuebingen_Neckarfront.jpg) | ![style](./img/vangogh_starry_night.jpg) |

| model | transferred image | runtime | remark |

| :-: | :-: | :-: | :-: |

| vgg19         | ![vgg19](./out/vgg19.jpg)                 |  39.086s | nice! |

| sd-vae-ft-mse | ![sd-vae-ft-mse](./out/sd-vae-ft-mse.jpg) | 213.099s | just ok |

| sdxl-vae      | ![sdxl-vae](./out/sdxl-vae.jpg)           | 213.138s | loss very large |

| taesd         | ![taesd](./out/taesd.jpg)                 |  35.035s | bad texture |

| taesdxl       | ![taesdxl](./out/taesdxl.jpg)             |  34.599s | bad texture |

### Quickstart

- install PyTorch-GPU follow the [official guide](https://pytorch.org/get-started/locally/)

- `pip install -r requirements.txt`

- run the demo images: `python run_.py`

- run your own images: `python run_.py -C /path/to/content.jpg -S /path/to/style.jpg`

### references

- 图像风格迁移详解: [https://zhuanlan.zhihu.com/p/55948352](https://zhuanlan.zhihu.com/p/55948352)

- https://github.com/EliShayGH/deep-learning-style-transfer

- https://github.com/gsurma/style_transfer

----

by Armit  

2024/02/13 init repo  

2024/10/08 merge new repo