Intel details ExtraSS framework featuring frame extrapolation technology

[harlan4096]

Intel proposes ExtraSS framework with frame extrapolation technology

Intel might be gearing up to enter the ongoing competition in enhancing image quality and boosting performance in modern graphics. In a recent research paper from SIGGRAPH Asia 2023 in Sydney, Intel is introducing a new approach, a framework that is based on frame extrapolation.



Wccftech have done a great job of explaining the paper called “ExtraSS: A Framework for Joint Spatial Super Sampling and Frame Extrapolation”. This paper introduces a combination of two key technologies based on spatial upscaling and frame extrapolation. The word extrapolation has an important difference when discussing Intel’s technology, because AMD FSR3 and DLSS3 are both based on frame interpolation. This key difference means that instead of relying on old and new rendered frames, the Intel method relies solely on old frames. This, of course, has some benefits and drawbacks.

The paper discusses the challenges of achieving real-time rendering with high-quality results. It explains the concepts of spatial and temporal super sampling and highlights existing methods like DLSS, XeSS, and FSR. The paper from Intel’s research proposes a unified framework that combines spatial upscaling and frame extrapolation to challenge these technologies.



As Intel notes, frame interpolation adds extra latency for the rendering pipeline. For this reason, NVIDIA and AMD methods rely on technologies such as Reflex or Anti-Lag+. In Intel’s case, that may not be necessary. Frame generation relying solely on prior frames will result in artifacts, but there are methods that use motion vectors combined with neural networks to improve shading and limit artifacts, but those are not perfect either. For this reason, Intel researchers are proposing a framework that uses both spatial super sampling and frame extrapolation together while staying efficient and high quality.
...

Continue Reading