Sony has filed a patent for an AI-driven system designed to stabilize frame rates in video games. The technology generates the next frame while the GPU is still rendering the current one, aiming to eliminate the latency issues common in traditional frame interpolation methods.
The system utilizes a pair of encoder and decoder models. The encoder converts the current frame into a compact numerical representation, while the decoder reconstructs the predicted image. To improve accuracy, the model incorporates motion vectors and player controller input data.
Training involves sequences of video frames where the neural network predicts the subsequent image and compares it against the actual frame to calculate errors. The patent also describes a method where the decoder produces additional color channels to generate multiple frames simultaneously.
An alternative approach mentioned in the patent involves a non-AI algorithm that shifts pixel blocks based on movement. While the documentation claims the system can maintain stable frame rates without sacrificing quality, the technology remains at the patent stage with no confirmed implementation.
