Two-Branch Neural Network for Real-Time Semantic Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern semantic segmentation methods in computer vision are inefficient and unable to perform real-time object classification, which is a critical requirement for applications like autonomous driving, despite achieving high accuracy.
Innovation Solution
A two-branch neural network architecture is employed, where a low-resolution branch processes context and a high-resolution branch refines boundaries, utilizing separable convolution modules to reduce computational cost and achieve real-time performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If modern semantic segmentation methods use deep convolutional neural networks with many operations and parameters, then segmentation accuracy is improved, but processing speed deteriorates (taking more than a second per image even on high-end GPUs)
Solution Approach 1:
The patent applies segmentation by dividing the image processing task into two separate branches: a low-resolution branch for global context understanding and a high-resolution branch for local detail processing. This segmentation allows each branch to be optimized independently, with the low-resolution branch using fewer operations for overall scene understanding and the high-resolution branch focusing computational resources only where detailed segmentation is needed, thereby improving processing speed while maintaining accuracy.
Solution Approach 2:
The patent implements local quality by applying different processing depths and resolutions to different regions of the image. The high-resolution branch processes only specific regions requiring detailed segmentation with more computational operations, while other regions are handled by the more efficient low-resolution branch. This selective application of computational resources maintains segmentation accuracy in critical areas while reducing overall processing time.
2Measurement precision
If semantic segmentation methods process images at high resolution to maintain boundary detection accuracy, then segmentation precision is improved, but computational cost increases
Solution Approach 1:
The patent segments the computational task by resolution levels, processing the entire image at low resolution for global context and only processing specific regions at high resolution for boundary detection. This reduces the total number of high-resolution operations from processing the entire image, thereby lowering computational cost while maintaining boundary detection accuracy where needed.
Solution Approach 2:
The patent applies partial action by performing high-resolution processing only on portions of the image that require detailed boundary detection, rather than processing the entire image at high resolution. The low-resolution branch handles the majority of the image area, and only regions identified as requiring detailed segmentation are processed at high resolution, reducing overall computational cost while maintaining necessary precision.
Data Source
AI summary
An image processing method for segmenting an image, the method comprising:receiving first image;producing a second image from said first image, wherein said second image is a lower resolution representation of said first image;processing said first image with a first processing stage to produce a first feature map;processing said second image with a second processing stage to produce a second feature map; andcombining the first feature map with the second feature map to produce a semantic segmented image;wherein the first processing stage comprises a first neural network comprising at least one separable convolution module configured to perform separable convolution and said second processing stage comprises a second neural network comprising at least one separable convolution module configured to perform separable convolution; the number of layers in the first neural network being smaller than the number of layers in the second neural network.