The patents, which were originally filed in March 2020 but were published on Thursday, detail “systems and methods for machine learned image conversion”.
It describes the process of taking a source image, dividing it into blocks of pixels and giving each block some context data. These are all then added to an ‘activation matrix’, which is run through a trained neural network to convert the image to a higher resolution. [...]
One example given details a process similar to the Switch’s main functionality, where a game can run at one resolution while a device is being powered by a battery, but when plugged into a power socket or connected to a TV the output can be upscaled to 1080p.
Another example discusses the process’s potential use in cloud gaming, in which cloud-based systems can send a compressed, low-res image to the player’s device, which will then use AI to upscale it to 1080p at the user’s end to reduce latency. [...]
It states: “It will be appreciated that while the example shown […] relates to transforming a 540p image to a 1080p image, the techniques herein may be applied to other image sizes (e.g. 720p to 1080p, 480p to 1080p, 1080p to 1440p, 1080p to 4K, 720p to 4k, etc).” [...]
Reply With Quote