AI image restoration model fixes old photos for free
Artificial Intelligence (AI) is making waves in the imaging world. The DALL-E Image Synthesizer recently entered beta testing. Google researchers just released a report on their AI imaging model called “Imagen,” which produces results that are frighteningly accurate. AI can even help photographers tag, organize, and edit photos. And now a new AI model called “Prerequisite of the generative face” (GFP-GAN) is available to restore your damaged old photos, with surprisingly good results…for free.
Related: Outsmart your iPhone’s overzealous camera AI
How to recover family photos using GFP-GAN
If you have old family photos that could use a little work, this AI image restoration template is available to the public right now. Techies can download the code herebut everyone is welcome to try the demo website. Simply upload a photo and watch the results. This can save you the tedious (and expensive) restore process.
How This AI Image Restoration Model Works
The document, titled “Towards real-world blind face restoration with a generative facial prerequisitepresents research showing a marked improvement in image regeneration using the Generative Facial Prior (GFP) model alone. But this particular model is assisted by Generative Adversarial Networks (GANs), so it is particularly good compared to other models. The PULSE modelfor example, does not use GFP and results tend to be low fidelity to the original image.
The GAN network is trained to generate images that produce accurate results. However, the addition of GFP provides feedback at different stages of the process to the generative model, helping to preserve identity and precise characteristics. This is why GFP-GAN has a reputation for making such faithful reproductions.
According Louis Boucharda Canadian AI/computer vision masters student who writes extensively on AI (although he is not the author of the study), GFP-GAN “will help their image restoration model to better match the features at each stage using this prior information from a powerful pre-trained StyleGAN-2 model known to create meaningful encodings and generate accurate images.This will help the model achieve realistic results while maintaining high fidelity .
As Bouchard and the researchers note, however, there is a downside in that there may be a “slight identity shift” in the results. This is because to be 100% accurate, the model would need a lot of information about what the person looked like in real life. Unfortunately, this is not always possible when working on damaged or low resolution images.
Although it’s not perfect, we have to admit that GFP-GAN does a pretty decent job. And, considering that it can help restore family photos (and other important photos) for free, it’s definitely worth a shot.