This AI-powered GFP-GAN tool lets you recover old damaged photos
Before the advent of smartphones and digital cameras, people used film cameras to record special occasions. The only issue with this is that an older photo has a higher chance of becoming ripped, cracked, or otherwise damaged. Family members who only have a few old photos of departed loved ones or distant ancestors they have never met may start to worry if the degradation is considerable. Now there seems to be a great solution out there to this problem, which of course involves AI (otherwise I would not write about it 🤣)
According to the online science site ZME Sciene, Chinese researchers have developed a new, potent AI technology that can aid in restoring the old photograph. The incredible computer programme uses a neural network known as the Generative Facial Prior-Generative Adversarial Network, or GFP-GAN for short, to eliminate creases, blemishes, grain, and other obvious signs of ageing from old photographs and restore them to as sharp as they would have been brand new.
Even with only a low resolution photograph, the effects seem truly incredible.
Let’s check out how everything functions first, though! These networks are algorithmic architectures that categorise examples as either “real” or “false” by pitting synthetic data against two neural networks: a so-called “generative model” intended to create new examples and a so-called “adversarial model” that categorises examples as either “real” or “false.” in order to produce anything that can be used as actual data.
This kind of modelling makes it possible to automatically find and learn from patterns in the input data to create results that are almost identical to the original. Additionally, it implies that the original image is sharpened and the visible explosion indicators are compensated for by the new artificial intelligence in the restored photos. The programme creates pixels that were never there, making this all a type of model fabrication. The conclusion is that the new technology accurately replicates the crucial facial features of the lips and eyes, which are crucial for maintaining identification.
The best part is, that it is available to everyone to try and use, for FREE! 🎉
You can just simoply upload your pictures and test it. I have tried it and indeed it works, although it did not work for every kind of image as good as some ‘advertised’ but that’s maybe because it needs a bit more tweeking.