Sony Patents a DLSS-like Machine Learning Image Reconstruction Technology for PlayStation 5

Sony Patents a DLSS-like Machine Learning Image Reconstruction Technology

Sony Patents a DLSS-like Machine Learning Image Reconstruction Technology

Sony has patented a machine learning algorithm which could deliver the console manufacturer higher fidelity visuals at a lower performance cost, using image reconstruction to provide similar results to Nvidia’s DLSS tech. 

The future of gaming will not be focused on pure resolution numbers. With techniques like Temporal upsampling, checkerboard rendering and DLSS 2.0, we already know that high-quality images can be produced with non-native pixel counts, and as reconstruction technologies continue to improve, native rendering will quickly become irrelevant. 

With DLSS 2.0, Nvidia has already shown us the potential of Machine Learning enhanced image reconstruction techniques, enabling higher performance levels with image quality levels that can sometimes exceed native resolution rendering. With this in mind, it is easy to see why game developers, console manufacturers and publishers are investing in similar techniques. 

Unfortunately, the full patent is only available in Japanese, but here is an English abstract from the document. This patent was uncovered by zerobfchallange on Reddit. 

 
An information processing device for acquiring a plurality of reference images obtained by imaging an object that is to be reproduced, acquiring a plurality of converted images obtained by enlarging or shrinking each of the plurality of reference images, executing machine learning using a plurality of images to be learned, as teaching data, that include the plurality of converted images, and generating pre-learned data that is used for generating a reproduction image that represents the appearance of the object.

  
Reconstruction techniques allow game developers to increase the graphical quality of their games when upscaling their titles from lower than native resolutions. If reconstruction techniques become effective enough, gamers will find it difficult to tell the difference between reconstructed images and native images. Reconstruction will allow developers to focus the performance of gaming hardware in new directions, making games run faster and look better. 

If Sony can create an effective image reconstruction solution, it will effectively give their PlayStation 5 hardware a significant performance upgrade, allowing more games to target 4K, higher framerates or both. 

Sony Patents a DLSS-like Machine Learning Image Reconstruction Technology  

You can join the discussion on Sony’s patented DLSS-like image reconstruction technique on the OC3D Forums.