Image Segmentation Neural Network Context-Switchable Depth Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for image segmentation using convolution neural networks lose valuable information from depth data, resulting in poor accuracy.
Innovation Solution
A method involving a full convolution neural network and a context-switchable neural network is employed, where the output from the full network is processed by the context-switchable network to generate intermediate characteristic patterns for image segmentation, utilizing hyperpixel districts and adjusting depth values to refine or extend context expression information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If depth data is encoded into three-channel images and processed by convolution neural network, then the network can be trained to calculate segmentation characteristics, but most useful information from depth data is lost in the output
Solution Approach 1:
The patent segments the processing of depth data by creating separate processing paths: one for RGB images and one for depth images. The depth image is processed through a dedicated context-switchable neural network that preserves depth information, while the RGB image is processed through a full convolutional network. This segmentation allows each data type to be processed with appropriate methods that preserve their unique characteristics.
Solution Approach 2:
The patent introduces an intermediary module that fuses the characteristic patterns from both RGB and depth processing paths. This intermediary fusion mechanism combines the segmentation characteristics from the full convolutional network with the context expression information from the context-switchable network, ensuring that depth information is not lost but integrated meaningfully into the final segmentation result.
2Measurement precision
If a single convolution neural network is used for image segmentation, then the processing is simple, but the segmentation accuracy is poor due to information loss
Solution Approach 1:
The neural network structure is segmented into two parallel pathways: a full convolutional network for RGB image processing and a context-switchable neural network for depth image processing. This segmentation allows each pathway to be optimized for its specific data type while maintaining manageable complexity through modular design.
Solution Approach 2:
The patent creates a multi-functional neural network system where the full convolutional network handles RGB image segmentation while the context-switchable network handles depth image processing. Both networks feed into a common fusion mechanism that produces the final segmentation result, making the overall system universal in handling multiple data types.
Data Source
AI summary
Methods for image segmentation, computer devices, and storage mediums. The method includes acquiring a to-be-segmented image, inputting the to-be-segmented image into an input variable of a full convolution neural network and outputting a convolution characteristic pattern; inputting the convolution characteristic pattern into an input variable of a context-switchable neural network and outputting context expression information; and generating an intermediate characteristic pattern for image segmentation according to the convolution characteristic pattern and the context expression information.