This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
This paper presents a portrait style transfer method that generalizes to various domains and enables high-quality semantically aligned style transfer for areas such as hair, eyes, eyelashes, skin, lips, and background. To this end, we obtain a distorted reference that is semantically aligned with the input by establishing a dense semantic correspondence between a given input and a reference portrait based on a pre-trained model and a semantic adapter. To ensure efficient and controllable style transfer, we devise the AdaIN-wavelet transform that balances content preservation and style transfer by blending the low-frequency information of the distorted reference with the high-frequency information of the input in the latent space. We also design a style adapter that provides style guidance from the distorted reference. Using the stylized latent space obtained from the AdaIN-wavelet transform, we generate the final result using a biconditional diffusion model that integrates ControlNet, which records the high-frequency information and style guidance. We demonstrate the superiority of the proposed method through extensive experiments. The code and trained models are available at https://github.com/wangxb29/DGPST .
