Skip to content
/ HS-Diffusion Public

HS-Diffusion: Semantic-Mixing Diffusion for Head Swapping

81 stars 4 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

qinghew/HS-Diffusion

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
Figures
Figures
 
 
README.md
README.md
 
 

Repository files navigation

HS-Diffusion

HS-Diffusion: Semantic-Mixing Diffusion for Head Swapping

Qinghe Wang, Lijie Liu, Miao Hua, Pengfei Zhu, Wangmeng Zuo, Qinghua Hu, Huchuan Lu, Bing Cao

This paper aims to stitch a source head to another source body, while maintaining the main components of the two source images unchanged. We first propose a semantic-mixing diffusion model for head swapping, which blends the semantic layouts to guide the mixing of diffusion latents step-by-step, stitching one head to another body seamlessly. We also propose a semantic calibration strategy to adaptively inpaint incomplete region and address the occlusion and noise issues encountered for head swapping.

Training process.

Inference process.

Results

Comparsion for head swapping:

Semantic-guided head replacement:

Semantic-guided local & multi-component replacement:

Head swapping in the wild:

Databases

We process the Stylish-Humans-HQ(SHHQ) dataset to the half-body SHHQ dataset as introduced in our paper. It can be downloaded from Baidu Drive with password mw4y.

TODOs

  • Release training data
  • Release training code
  • Release inference code

Contact

If you have any questions or suggestions about the paper, feel free to reach me ([email protected]).

About

HS-Diffusion: Semantic-Mixing Diffusion for Head Swapping

Resources

Readme
Activity

Stars

81 stars

Watchers

33 watching

Forks

4 forks
Report repository

Releases

No releases published

Packages

No packages published