An open source tool is available on GitHub with which faces can be recovered from old photos: Generative Facial Prior (GFP) is the name of the tool developed in China and is a Generative Adversarial Network (GAN) type machine learning model. According to the project description, GFP-GAN provides “practical algorithms for realistic reconstruction of faces” in images.
GFP-GAN is open source software licensed under the Apache 2.0 license and comes from a development team at ARC Lab, where ARC stands for Applied Research Center. Behind him is Chinese social media provider Tencent, who founded the lab in 2019 and claims to be conducting media-related AI research with it.
Under the Hood: image generation in several stages
Canadian Computer scientist and master’s student in artificial intelligence Louis Bouchard He took a closer look at the underlying technology and presented the tool on his YouTube channel “What is AI”: Whereas traditional methods for recovering old photos have so far used an AI model that measures the differences between the created photos and the original, the new technology apparently collects information from Two complementary AI models and adds realistic missing details.
According to Bouchard, the new approach uses a pre-trained version of the AI model, which divides the image creation process into several stages. Thanks to technology, the identity of people in photos can be preserved better than before – among other things, because special attention is paid to facial features such as the eye and mouth area. New technology is also not perfect, so especially old or damaged photos are given new details that were not there in the original. Depending on the state of preservation, the reconstructed representations can look significantly different from the people in the original.
GitHub repository and demos are available online
There is a Colab demo of the project as well as online demos at Huggingface, Replicate, and BaseTen, among others. If you want to run GFP-GAN yourself, you need Python version 3.7 or higher, or PyTorch version 1.7 or higher. The ARC Lab team also recommends installing Anaconda or Miniconda. Optionally, the model can be run with NVIDIA GPU and CUDA, and it should be possible to run under Windows as well as under Linux. The GitHub repository provides installation notes, training tips, and quick inference.
If you want to try the tool, you can Load images for optimization in the browser via a web application called BaseTen or get Download the source code from the Github repository and integrate the model into your own applications.
“Subtly charming coffee scholar. General zombie junkie. Introvert. Alcohol nerd. Travel lover. Twitter specialist. Freelance student.”