Feature Map Convolution Using Parallel Anchor Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing convolution context models in image processing suffer from high time complexity and slow convolution speed due to the need for sequential updating of feature points, reducing image processing efficiency.
Innovation Solution
A feature map processing method that determines the positions of multiple convolutional kernels simultaneously, allowing for parallel convolution operations based on the positions already updated by previous kernels, thereby reducing the need for serial processing and improving efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sequential updating of feature points is used in convolution context models, then feature map processing can be performed, but time complexity increases and convolution speed decreases
Solution Approach 1:
The patent segments the feature map processing into multiple independent convolution operations that can be performed in parallel. Specifically, it divides the processing into: (1) first convolution operation on the input feature map, (2) second convolution operation on the updated feature map, and (3) third convolution operation on the further updated feature map. These segmented operations can be executed simultaneously, breaking the sequential dependency and reducing overall processing time while maintaining correctness.
Solution Approach 2:
The patent applies preliminary action by pre-calculating and storing the positions that convolutional kernel anchors have already passed during the scanning process. This preliminary information is then used to determine the positions for subsequent convolution operations, allowing the system to skip redundant calculations and directly proceed to the next necessary convolution step, thereby reducing time complexity.
2Manufacturing precision
If sequential convolution operations are performed, then feature points are updated correctly, but processing efficiency is reduced
Solution Approach 1:
The patent segments the convolution processing into multiple independent operations that can be parallelized. By dividing the feature map processing into separate convolution operations (first, second, and third convolutions) that can be performed simultaneously on different regions or stages, the system maintains precise feature point updating while significantly improving processing efficiency through parallel execution.
Solution Approach 2:
The patent introduces a new dimension of parallelism by performing multiple convolution operations simultaneously rather than sequentially. It uses the already-passed anchor positions to identify independent operation opportunities, effectively transforming the time dimension into a parallel execution space where multiple convolutions occur at the same time, thus improving productivity without sacrificing accuracy.
3Loss of time
If parallel convolution operations are implemented, then processing speed increases, but position determination complexity increases
Solution Approach 1:
The patent reduces position determination complexity by performing preliminary action: it pre-calculates and stores the positions that convolutional kernel anchors have already passed during the scanning process. This preliminary information is cached and reused to determine positions for subsequent parallel convolution operations, avoiding redundant calculations and simplifying the position determination process despite the increased parallelism.
Solution Approach 2:
The patent uses copying by replicating the position determination logic across multiple convolution operations. Instead of independently calculating positions for each parallel operation, it copies the already-passed anchor position information and uses it to derive positions for subsequent convolutions, reducing the overall computational complexity while enabling parallel execution.
Data Source
AI summary
The present disclosure relates to a feature map processing method and apparatus, and a computer-readable storage medium, and relates to the field of image processing. The feature map processing method includes: determining a position of an anchor of a first convolutional kernel in a feature map at a next moment, according to a position of the anchor of the first convolutional kernel in the feature map at a current moment; determining a position of an anchor of at least one second convolutional kernel in the feature map at the next moment, according to a position that the anchor of the first convolutional kernel has already passed; and performing convolution on the feature map region corresponding to the first convolutional kernel by the first convolutional kernel and performing convolution on the feature map region corresponding to the at least one second convolutional kernel by the at least one second convolutional kernel simultaneously, at the next moment.