Sony patent targets AI frame trimming for streamed games
The filing describes a sender-side network that scores image sections and strips those scoring below 0.4, with the receiving device rebuilding detail.
Sony published a patent on October 29 describing an AI-based method for cutting down the image data required to stream video, according to Patentlyze. The filing sets out a system that discards parts of each frame on the sending side and reconstructs them on the device receiving the stream.
At the centre of the patent is a decimation network that divides an image into a grid and assigns each section a score. Portions that score below 0.4 are removed before transmission instead of being sent in full.
A reconstruction network on the receiving device then fills the discarded areas using anchor elements, restoring image detail at the other end of the connection.
The system can also take a target image quality or compression level as an input. That setting determines how many portions of the frame are removed and reconstructed, tying the amount of data sent to the quality level requested.
Background
The patent arrives during a period of rising hardware prices. As of recent months, the PlayStation 6 and Xbox Project Helix have been expected to cost significantly more than their predecessors, and the PlayStation Portal has proven more popular than Sony anticipated.
Quick answers
What does Sony's patent describe?
An AI method that splits streamed images into a grid, removes sections scoring below 0.4, and reconstructs them on the receiving device.
When was the patent published?
Sony published the patent on October 29, according to Patentlyze.
Can the system target a specific quality level?
Yes. It can be given a target image quality or compression level, which sets how many portions are removed and reconstructed.