A neural network-based real-time energy consumption optimization drawing method and device

Abstract

The present invention discloses a neural network-based real-time energy consumption optimized drawing method and device, comprising: acquiring drawing-related data sets, including GPU energy consumption, drawing parameters and data related to GPU energy consumption, drawing parameters and data related to image quality ; According to the drawing related data set, the energy consumption prediction model and the error prediction model are constructed based on the neural network, and the energy consumption prediction model and the error prediction model are used to perform the energy consumption screening based on the GPU energy consumption budget threshold and the image error prediction threshold based on the image error prediction model. Error screening to determine the best rendering parameters to be selected, and then use the best rendering parameters for real-time rendering of the 3D scene. In this way, while ensuring the image quality as much as possible, it significantly reduces the energy consumption during the drawing process and prolongs the battery life.

Description

technical field [0001] The invention belongs to the field of real-time rendering, and in particular relates to a neural network-based real-time energy consumption optimization rendering method and device. Background technique [0002] Rendering is a process of converting a three-dimensional geometric model into a two-dimensional image. Drawing a 3D animation scene is a very time-consuming process, because an animation generally consists of tens of thousands of frames of images. As people's requirements for visual effects increase, the resolution of each frame of image is getting higher and higher, and there are more and more additional visual effects, and the computing resources required to draw a frame of image are also increasing rapidly. [0003] With the continuous improvement of modern graphics hardware acceleration technology, the rapid development of graphics processing unit (GPU) has greatly improved the speed and quality of computer graphics processing, and promote...

AI Technical Summary

Problems solved by technology

Because these methods are implemented by the underlying hardware, it is necessary to change the hardware structure of the existing equipment, which has the characteristics of high cost and low versatility.

[0007] Recently, although Zhang et al. proposed an effective energy consumption optimization solution based on energy consumption-error prediction and budget in "On-the-Fly Power-Aware Rendering" and an invention patent with the announcement number CN109191555B disclosed a method based on Real-time drawing method of energy consumption-error prediction and budget, but there are still many problems: when the GPU operating frequency changes greatly, the energy consumption prediction result is not accurate enough, the real-time power of the GPU needs to be measured during drawing, and frame-by-frame cannot be achieved Energy consumption optimization, large resource overhead for predicting image errors, etc.

Images