Valid Region Pixel Data Writing for Memory Bandwidth Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large amounts of pixel data from input images occupy significant memory and bandwidth, leading to decreased operation efficiency in multimedia and artificial intelligence systems.
Innovation Solution
A data processing method that defines a valid region within the input image and writes only valid pixel data to specific target addresses in memory, using address allocation relations based on the starting write address, valid width, and accumulation address to optimize memory usage and reduce bandwidth access.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all pixel data from input image is written to memory, then complete image data is stored, but memory usage and bandwidth consumption increase significantly
Solution Approach 1:
The patent extracts only the valid pixel data from the input image based on a predefined valid region, excluding invalid or unnecessary pixels. This extraction principle directly reduces the quantity of data written to memory while preserving the essential image information needed for processing, thereby resolving the contradiction between data completeness and memory usage.
Solution Approach 2:
The patent applies different quality standards to different regions of the image by defining a valid region where only pixels within this region are written to memory. This local quality approach allows the system to maintain high data quality in critical areas while reducing overall data volume, effectively balancing data completeness with memory efficiency.
2Productivity
If all pixel data from input image is accessed in memory at high frequency, then complete image processing is enabled, but bandwidth consumption increases and operation efficiency decreases
Solution Approach 1:
The patent extracts only the valid pixel data that needs to be processed, eliminating unnecessary data access operations. By writing only valid pixels to memory based on the valid region definition, the system reduces the frequency and volume of memory access, thereby lowering bandwidth consumption while maintaining full processing capability for the essential image data.
3Quantity of substance
If valid region is defined and only valid pixel data is written to memory, then memory usage and bandwidth consumption are reduced, but data processing complexity increases
Solution Approach 1:
The patent performs preliminary action by pre-defining the valid region and identifying valid pixels before the actual data writing process. This preprocessing step, which includes determining the valid region boundaries and generating the valid pixel mask, simplifies the subsequent memory writing operation by providing clear guidance on which pixels to write, thereby reducing overall processing complexity despite the added initial step.
4Speed
If address allocation is performed according to starting write address and valid width, then memory access efficiency is improved, but address calculation complexity increases
Solution Approach 1:
The patent changes the addressing parameters from standard sequential addressing to a customized addressing scheme based on starting write address and valid width. This parameter change enables more efficient memory access by aligning the addressing with the actual valid data layout, improving memory access speed while the added calculation complexity is offset by the performance gains and the use of systematic address allocation relations.
Data Source
AI summary
A data processing method includes the following steps. An input image is received, the input image has a plurality of pixels, and each of the pixels has a pixel position. An valid region is defined in the input image, a plurality of valid pixels among the pixels are located in the valid region, and each valid pixel has a valid data. The valid data is written to the target address of the memory correspondingly according to a starting write address of the memory and a valid width of the input image, or according to the starting write address, the starting offset of the input image and the accumulation address of the memory.