If you’re a fan of older films, you’ll be all to familiar with grainy, murky footage, and with today’s high standards for realism, that simply isn’t acceptable. To breathe new life into older movies and clips, Denis Shiryaev turned to modern-machine learning to upscale the classic film to modern day 4k 60fps standards.
He managed to achieve this by using a software called Gigapixle AI, created by Topaz labs which allowed him to upcast images by 600 percent. To prevent the upscaled image from looking too blurry, Gigapixel AI then used advanced neural networks to add realistic details into an image.
Basically, if you show the neural network enough examples of a face, and then show it a poor quality image of someone face, it will then be able to fill in the right details for the subject, such as eyes, nose, and mouth. It works in the same manner as regards to backgrounds. If you show a neural network a low-resolution image of a building, it will know to fill in minute details bricks and windows.
“A neural network analyses thousands of photo pairs to learn how details usually get lost,” Topaz Labs explained on their product page for Gigapixel AI. “The algorithm learns to ‘fill in’ information in new images based on what it has learned, effectively adding new detail to your photo.”
The final video is by no means a perfect recreation of the original video. Upon closer inspection you can see evidence multiple visual artifacts popping in and out of frame like the rippling train motion and melding pedestrians, but if Shiryaev is able to perfect his technique we could see a slew of older movies making a high-resolution comeback.