Visual Perception-Driven Dynamic Adaptive Shading

This paper proposes an algorithm that dynamically and adaptively adjusts the rendering shading rate based on pre-rendered low-resolution images. We first pre-render the current frame at low resolution using full shading rate and various low shading rates. Next, we use the difference between low shading rate and full shading rate pixel colors to determine regions where using a low shading rate will not have an impact. At the same time, we obtain the saliency map of the current frame through a simplified visual saliency model. Finally, the saliency map and the pixel color difference map are comprehensively analyzed to generate the shading rate image of the current frame. Our implementation uses computed per-tile shading rates with variable rate shading to lower shading costs. Experimental results show that our method can improve rendering performance with almost no visual quality loss.