Streaming Image Scaling Structure for CPU-Free AI Preprocessing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The use of CPU for resizing images in traditional calculation architectures leads to high resource occupation and reduced end-to-end performance of AI systems, increasing calculation time and affecting overall efficiency.
Innovation Solution
An image size adjustment structure and method based on a streaming architecture, utilizing multiplication and addition operation units with data registering and selection units, allows image data to flow through these units sequentially, bypassing CPU calculations and employing bilinear or cubic interpolation algorithms for efficient scaling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If CPU is used to resize images in traditional calculation architecture, then image resizing can be performed, but CPU resources are heavily occupied and calculation time increases
Solution Approach 1:
The patent replaces the traditional CPU-based image resizing mechanism with a dedicated hardware resizing unit that implements interpolation algorithms (bilinear or cubic) in hardware. This substitution moves the resizing function from software execution on a general-purpose processor to specialized hardware circuitry, enabling parallel computation and significantly reducing calculation time while improving productivity.
2Productivity
If CPU is used for image resizing, then image scaling can be achieved, but overall AI system performance is reduced due to high resource occupation
Solution Approach 1:
The patent segments the AI system into distinct functional modules: a dedicated image resizing unit with multiplication and addition operation units for scaling operations, and the neural network processing unit. This segmentation allows the resizing function to be handled by specialized hardware, reducing CPU resource occupation and enabling the main processor to focus on neural network computations, thereby improving overall system productivity without excessive complexity.
3Adaptability or versatility
If traditional calculation architecture is used, then existing systems can maintain compatibility, but end-to-end efficiency of streaming architecture AI chip is reduced
Solution Approach 1:
The patent introduces an intermediary image resizing unit that acts as a bridge between the streaming architecture AI chip and the neural network processing unit. This intermediary component handles preprocessing of image data through hardware-accelerated resizing operations, ensuring compatibility with various input formats while optimizing data flow efficiency for the streaming architecture, thereby improving end-to-end productivity without sacrificing adaptability.
Data Source
AI summary
The present application discloses an image size adjustment structure, an adjustment method, and an image scaling method and device based on a streaming architecture. The image size adjustment structure includes: a first multiplication operation unit, a second multiplication operation unit, a first data registering unit, a second data registering unit, a first addition operation unit and a second addition operation unit, and the input and output ports of each unit are connected according to the specified data flowing direction. It realizes the fast calculation of image data and relieves the calculation pressure of CPU.