CNN Feature Map Selection for Faster Image Correspondence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image correspondence techniques, such as HOG, SIFT, and CNN, struggle with efficiency and accuracy in establishing correlations between images, particularly in dynamic feature selection and global transformation estimation.
Innovation Solution
A method and apparatus using a convolutional neural network (CNN) to dynamically select relevant feature map pairs based on neural networks, generating hyper feature maps through upsampling and concatenation, and calculating correlations to improve image correspondence accuracy and speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all feature map pairs from multiple CNN layers are used for image correspondence, then accuracy is improved, but computational complexity and processing time increase
Solution Approach 1:
The patent segments the feature map pairs from multiple CNN layers into two groups: relevant feature map pairs that contribute to accurate correspondence, and irrelevant ones that can be discarded. This segmentation allows the system to process only the necessary feature maps, reducing computational complexity while maintaining accuracy.
Solution Approach 2:
The patent extracts and selects only the relevant feature map pairs from the complete set of feature maps generated by multiple CNN layers. By taking out only the essential feature maps needed for accurate correspondence, the system avoids the computational burden of processing all feature maps while preserving accuracy.
2Measurement precision
If all feature map pairs from multiple CNN layers are used for image correspondence, then accuracy is improved, but processing speed decreases
Solution Approach 1:
The patent segments the feature map pairs into relevant and irrelevant categories, processing only the relevant portion. This segmentation enables the system to achieve accurate correspondence results while significantly reducing processing time by excluding unnecessary computational steps.
Solution Approach 2:
The patent applies partial action by processing only a subset of feature map pairs that are determined to be relevant, rather than processing the complete set. This partial processing approach maintains accuracy while improving processing speed by avoiding excessive computation on irrelevant feature maps.
3Loss of energy
If irrelevant feature map pairs are processed, then computational resources are wasted, but accuracy may be compromised if relevant features are excluded
Solution Approach 1:
The patent employs a feedback mechanism where the relevance of each feature map pair is evaluated based on its contribution to correspondence accuracy. This feedback allows the system to dynamically identify and select only the relevant feature map pairs, ensuring that computational resources are not wasted on irrelevant data while maintaining high accuracy.
Solution Approach 2:
The system performs self-service by automatically evaluating and selecting relevant feature map pairs without requiring external intervention or manual curation. The relevance determination is performed autonomously, ensuring efficient use of computational resources while preserving accuracy through intelligent feature selection.
Data Source
AI summary
A method with image correspondence includes: acquiring a plurality of feature map pairs corresponding to outputs of a plurality of layers of a convolutional neural network (CNN) in response to an input of an input image pair; selecting a portion of feature map pairs from among the plurality of feature map pairs based on a feature of each of the plurality of feature map pairs; generating a hyper feature map pair based on the selected portion of feature map pairs; and generating a correspondence result of the input image pair based on a correlation of the hyper feature map pair.