This is the official code of HQFace, which is implemented by PyTorch.
Abstract
we propose a high-quality face swapping framework (HQFace). Analysis of the latent space of HQFace revealed varying degrees of identity information in feature blocks at different levels. Building on this discovery, a multi-scale adaptive identity mapping module is proposed to locate the identity information of different feature levels for achieving identity information transformation while keeping other attributes unchanged. Furthermore, to further disentangle identity information from other information in latent codes, a new dual en-decoding generation strategy is proposed. This strategy enhances the authenticity of facial regions, particularly the hue of facial images.
Some results of HQFace
There are the some results of face reenactment realized by HQFace.
- To do list
- Upload the PyTorch source
- Presentation effect
- Edit how to use code
🔊 We will soon reveal how to use code to achieve high-quality face swapping!