Last week, my colleague Tristan wrote about an AI developer who used machine learning to upscale the famous 1895 train scene to 4K at 60 frames per second. While this was a great short watch, it made me wonder about using AI to restore and enhance old videos. Thankfully, I stumbled upon a paper featured by the Two Minute Papers YouTube channel over the weekend that aims to improve and colorize these videos. The model uses Temporal Neural Network to identify and correct defects such as flickers in vintage videos. The remastering of vintage films comprises of a diversity of sub-tasks including super-resolution, noise removal, and contrast enhancement which jointly aim to restore the deteriorated film medium to its original state.

A developer and YouTuber named Denis Shiryaev recently used "several neural networks" to update a famous French short film from 1895 to make it appear as though it were shot on a modern phone. Called "L'arrivée d'un train en gare de La Ciotat," the 50-second French short documentary shows a train arriving at Ciotat train station. The film is famous for allegedly causing theater goers at the time to panic out of fear the train would burst through the screen and crash into them. While this tale is largely dismissed as little more than legend, it's easy to imagine the impact such a visual would have had on someone who'd never seen a motion picture before. Shiryaev apparently felt the classic masterpiece needed a modern update so they employed what they refer to on Reddit as "several neural networks" to upscale the film to 4K resolution at 60 fps. The video's also had sound added, which makes it all the more eerily… normal.