CNN Training With Cross-Scale Feature Map Consistency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing convolutional neural networks struggle to detect the same features for image elements at different scales, leading to inefficiencies in tasks like object detection and semantic segmentation.
Innovation Solution
A method for training convolutional neural networks by processing scaled versions of input images and calculating a loss between feature maps generated by corresponding convolutional layers, ensuring feature correspondence across different resolutions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multi-scale networks with shared features are used, then the network can detect features at different resolutions, but the network creates reserved subspaces that prevent consistent feature detection across scales
Solution Approach 1:
The patent applies feedback by calculating a loss function that compares feature maps from different scales and uses this feedback to update network parameters. The loss function measures the discrepancy between features at different resolutions and feeds this information back to the training process, enabling the network to learn consistent feature representations across scales by minimizing the calculated loss.
2Adaptability or versatility
If additional convolutional layers are added to improve multi-scale detection, then feature detection capability improves, but computational resources and memory requirements increase
Solution Approach 1:
The patent makes existing convolutional layers serve multiple functions by having them process both original-resolution images and scaled-resolution images. The same layers are reused across different scales, eliminating the need for separate dedicated layers for each scale. This universal approach allows the network to detect features at multiple resolutions while maintaining the same computational resources and memory requirements.
Data Source
AI summary
A method for training a convolutional neural network. For each of a multiplicity of training input images, the method includes processing of the training input image by the convolutional network; processing of a scaled version of the training input image by the convolutional network; determining a pair of convolutional layers of the convolutional network so that a convolutional layer of the pair generates a first feature map for the training input image which has the same size as a second feature map which is generated by the other convolutional layer of the pair for the scaled version of the training input image; and calculating a loss between the first feature map and the second feature map; and training the convolutional neural network to reduce an overall loss which includes the calculated losses.