Backbone Network Generation for Intel CPU Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing lightweight backbone networks, such as SHuffleNetV2 and MobileNetV3, are not ideal for real-time predictions on Intel CPUs due to high inference times in tasks like target detection and image segmentation, failing to provide efficient performance for deep learning-based computer vision tasks.
Innovation Solution
A method for generating a backbone network that involves training and inferring initial backbone networks using training and inference images to determine a basic backbone network with optimal inference time and accuracy, and then enhancing it with a preset target network to improve inference accuracy without significantly increasing inference time, specifically optimized for Intel CPUs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If lightweight backbone networks such as ShuffleNetV2 and MobileNetV3 are used, then device complexity is reduced, but inference time increases and real-time prediction capability is lost
Solution Approach 1:
The patent changes the structural parameters of the backbone network by replacing standard convolutional layers with depthwise separable convolutions, adjusting channel dimensions, and modifying network depth to optimize for Intel CPU performance while maintaining lightweight characteristics
Solution Approach 2:
The backbone network is segmented into multiple stages with progressively increasing channel dimensions, allowing the network to process features at different levels of abstraction efficiently on Intel CPUs without excessive inference time
2Ease of manufacture
If existing lightweight backbone networks are used, then ease of manufacture is improved, but inference accuracy is insufficient for real-time predictions
Solution Approach 1:
The patent introduces dynamic channel attention mechanisms that adaptively adjust feature channel importance based on input characteristics, enabling the network to maintain high accuracy while preserving the lightweight structure for easy deployment on Intel CPUs
3Measurement precision
If backbone network accuracy is improved, then measurement precision is improved, but inference time increases
Solution Approach 1:
The patent extracts and removes redundant computational operations from the backbone network, keeping only the essential features needed for accurate prediction on Intel CPUs, thereby maintaining inference accuracy while reducing inference time
Data Source
AI summary
The present disclosure provides a method for generating a backbone network, an apparatus for generating a backbone network, a device, and a storage medium. The method includes: acquiring a set of a training image, a set of an inference image, and a set of an initial backbone network; training and inferring, for each initial backbone network in the set of the initial backbone network, the initial backbone network by using the set of the training image and the set of the inference image, to obtain an inference time and an inference accuracy of a trained backbone network in an inference process; determining a basic backbone network based on the inference time and the inference accuracy of the trained backbone network in the inference process; and obtaining a target backbone network based on the basic backbone network and a preset target network.