Tile-Based Rendering for Binocular Disparity Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Graphics processing units in portable devices face challenges in achieving graphics processing capabilities similar to workstations due to limited processing power and memory, making it difficult to efficiently render binocular disparity images.
Innovation Solution
A tile-based rendering method that involves performing tile binning on binocular disparity images using multiple graphics processors, determining the tile with the highest similarity between two images, and determining a rendering order to optimize tile rendering, thereby reducing memory access and improving performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional rendering methods are used on portable devices, then graphics processing capability can be maintained, but power consumption increases and processing efficiency decreases
Solution Approach 1:
The image is divided into multiple tiles, and rendering is performed independently on each tile. This segmentation allows the graphics processor to process smaller regions separately, reducing the need for frequent memory access and enabling more efficient use of on-chip memory, thereby improving processing efficiency and reducing power consumption on portable devices
Solution Approach 2:
Tile binning information is calculated in advance before actual rendering occurs. By pre-determining which tiles need to be rendered and their rendering order, the system avoids unnecessary processing and memory access during the actual rendering phase, improving overall processing efficiency and reducing power consumption
2Productivity
If tile rendering is performed without optimization, then rendering can be completed, but memory access efficiency remains low
Solution Approach 1:
Tile binning information is pre-calculated to determine the rendering order of tiles. This preliminary action allows the system to access and process tiles in an optimized sequence, reducing random memory access and improving memory access efficiency, thereby reducing memory access time and improving rendering efficiency
Solution Approach 2:
Tile binning information is stored and reused for determining rendering orders. By copying and reusing this pre-calculated information across different rendering operations, the system avoids redundant calculations and improves memory access patterns, reducing memory access time and enhancing rendering efficiency
3Productivity
If binocular disparity images are rendered simultaneously, then both images are processed, but processing complexity and memory usage increase
Solution Approach 1:
The rendering process is segmented into independent tile-level operations. By processing tiles from both left and right images separately using the same rendering pipeline, the system handles binocular disparity images efficiently without increasing overall processing complexity, maintaining high rendering throughput while managing complexity through modular tile-based processing
Solution Approach 2:
The rendering order of tiles is dynamically adjusted based on binning information and similarity metrics. By changing the processing parameters (rendering order) rather than the fundamental processing architecture, the system can efficiently handle binocular disparity images with reduced complexity while maintaining high throughput
Data Source
AI summary
A tile-based rendering method and a tile-based rendering apparatus are provided. The tile-based rendering method involves performing tile binning on a first image and a second image having binocular disparity using a plurality of graphic processors based on a draw command, determining a tile of the second image having a highest similarity to a tile of the first image based on a result of the tile binning, determining a tile rendering order for each tile of the first image and the second image, based on a result of the determining of the tile of the second image having a highest similarity to the tile of the first image, and performing tile rendering, according to the determined tile rendering order.