This repository contains the code and resources for the thesis project on generating and editing photo-realistic baby faces using transfer learning techniques with a pre-trained diffusion model (DM). The project explores generating high-quality baby faces conditioned by facial features, ethnicity, and expression through textual prompts. It also introduces novel pipelines to modify non-identity attributes like expressions and pose orientations.
- Our model achieves high realism, with 61.1% of participants unable to distinguish real baby faces from AI-generated ones.
- The approach demonstrates the effectiveness of DMs in generating and editing baby faces.
Generating realistic human faces is crucial for various applications such as advertising, film production, data augmentation, and medical imaging. However, generating realistic baby faces poses unique challenges due to the lack of comprehensive datasets and specific research in this area. Our project addresses these challenges by leveraging state-of-the-art generative models and transfer learning techniques.
- Generate realistic baby faces from textual descriptions.
- Enable modifications to various aspects of the generated faces, such as expressions and spatial orientation.
For face generation and edition purposes, we have developed a package that can be found at baby_face_generation_package. See a basic example in example.py. For the orientation modification pipeline, an example can be found at scripts/orientation-modification/ip-adapter-2.py.
This project is licensed under the MIT License. See the LICENSE file for more details.
For any questions or inquiries, please contact [email protected].
We hope this project provides valuable insights and tools for researchers and practitioners working on realistic face synthesis.