Welcome to Single-Image-Facial-Superresolution-GAN, a machine learning project developed as part of IIT CS 584. This project focuses on the fascinating realm of generating high-resolution face images from low-resolution inputs. The objective is to craft a controllable and efficient generative face super-resolution framework that not only produces high-fidelity face images with intricate details but also attains state-of-the-art performance.
-
Cutting-edge Super Resolution: Employing advanced techniques in deep learning and generative adversarial networks, this project offers an innovative solution to the challenge of generating high-resolution facial images from low-resolution inputs.
-
Controllable Framework: The proposed framework introduces an encoder-generator architecture, a style modulator, and a facial modulator. This allows for fine-tuned control over the generated images, enabling users to guide the super-resolution process according to desired characteristics.
-
Realistic Detail Generation: The framework excels at infusing generated face images with realistic details, enhancing the quality and authenticity of the output. This is achieved through careful design and utilization of neural network components.
-
End-to-End Training: The framework is designed to be end-to-end trainable, streamlining the training process and facilitating robust convergence. This design choice contributes to faster convergence times, enabling quicker iterations and experimentation.
-
Encoder-Generator Architecture: The backbone of the framework consists of an encoder-generator architecture, which forms the basis for converting low-resolution inputs into high-resolution outputs.
-
Style Modulator: The incorporation of a style modulator enhances the control users have over the visual appearance of the generated images, allowing for adjustments in attributes such as lighting, color balance, and more.
-
Facial Modulator: The facial modulator component focuses on the intricacies of facial features, enabling users to fine-tune aspects like facial expressions, eye gaze, and other nuances that contribute to the overall authenticity of the generated images.
To get started with the project, follow these steps:
- Clone this repository:
git clone https://github.com/yourusername/Single-Image-Facial-Superresolution-GAN.git - Explore the provided Jupyter notebooks to understand the training process, framework components, and examples of high-resolution image generation.
- Mohammad Firas Sada
- Antoine Courbot
- Aleksander Popovic
All source code included in this project is licensed under the GPL License.