Intention-Aware LOD: Predictive GPU Resource Allocation Method for Improving Perceptual Quality of 3D Object Rendering
In games, it is more important to improve the quality of 3D objects as perceived by players rather than the quality of rendering 3D objects faithfully to models. Thus, this invention proposes, as a method of improving the effective quality of rendering 3D objects as perceived by a player, a method of predicting an object that a player will next pay attention to and allocating more resources (shader units) that determine the rendering quality in descending order of the possibility of receiving attention. Here, the quality of rendering 3D objects refers to the number of polygons used in rendering a single object (LOD: Level of Detail) and the complexity of coloration processing by pixel shaders. Recent GPUs include a large number of programmable integrated shaders and realize higher rendering qualities by adjusting the roles of those shaders in accordance with the nature of each game. According to this invention, actions that can be taken next by a player are predicted by analyzing a huge operation history of that player, and objects that will be affected by the predicted actions and objects that serve as the executors of those actions are extracted as objects having high possibilities of receiving the player's attention. Furthermore, these objects are rendered by using a greater number of shaders as their possibilities of receiving attention become higher. Assuming that there are ten primary objects in a game screen and that the player can simultaneously pay attention to about one to three objects at most, by applying this invention to a game, it is possible to improve the rendering quality as perceived by a player by about 10% to 30% compared with before in the case where the same hardware is used (if it is acceptable to lower the rendering quality of each object not receiving attention by about 30%).
Full Text: Google Patents