Configurable Foveated Downsampling Filters for Lower CPU Bandwidth
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Image processing pipelines consume significant CPU bandwidth and resources, affecting overall compute resource utilization, particularly in devices with separate hardware components dedicated to image processing.
Innovation Solution
Implement configurable downsampling filters for foveated downsampling, determining downsampling phases, configuring data structures with interpolation coefficients, and dynamically generating filter types based on implementation-specific parameters to optimize resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If a separate image processing pipeline is implemented, then CPU bandwidth and computing burden are reduced, but overall compute resource utilization (memory, storage, power, I/O) is affected
Solution Approach 1:
The patent implements dynamic filter configuration where the downsampling filter type is determined based on the downsampling factor. Different filter types are selected dynamically depending on the specific downsampling requirements of different image regions, allowing the system to adapt to varying computational needs and optimize resource utilization without requiring a fixed, resource-intensive pipeline.
Solution Approach 2:
The patent changes the parameter of filter type selection based on downsampling factors. By configuring different filter types (e.g., first downsampling filter, second downsampling filter) according to the specific downsampling requirements, the system optimizes compute resource usage while maintaining processing efficiency in the separate pipeline.
2Ease of operation
If large data structures with all possible filter types are maintained, then filter selection is simple, but memory and storage resources are consumed
Solution Approach 1:
The patent extracts only the necessary filter types for each downsampling factor rather than maintaining all possible filter types in a single large data structure. Filter types are determined and applied selectively based on the specific downsampling requirements of different image regions, reducing memory consumption while maintaining ease of operation through direct mapping from downsampling factor to filter type.
Solution Approach 2:
The patent segments the filter selection process by creating separate data structures or lookup tables for different downsampling factors. Instead of a single monolithic data structure containing all filters, the system uses segmented organization where each downsampling factor has its own associated filter types, reducing overall memory footprint while simplifying filter retrieval.
Data Source
AI summary
Disclosed herein are a system, method, and computer program product embodiments for performing foveated downsampling based on configurable downsampling filters. For example, a number of downsampling phases between pixels of an image is determined. Based on the number of downsampling phases, a data structure that associates one or more interpolation coefficients with a downsampling filter type is configured. A downsampling factor for downsampling pixel values in a region of the image is determined. Based on the downsampling factor, a downsampling filter type from the data structure is determined. Pixel values in the region of the image are downsampled based on the one or more interpolation coefficients associated with the determined downsampling filter type.