Neural Network Object Prediction via Intermediate Result Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The accuracy of multiple target prediction results in deep learning technologies is low when predicting multiple targets simultaneously, due to significant differences between various prediction tasks such as depth estimation and scene segmentation, leading to poor prediction performance.
Innovation Solution
An object prediction method that involves feature extraction, intermediate prediction determination, fusion processing, and target prediction determination using a neural network, where feature information is extracted and multiple intermediate prediction results are fused to improve the accuracy of target prediction results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple targets are predicted simultaneously using a neural network, then the prediction process can be completed in one operation, but the accuracy of the obtained multiple target prediction results is low
Solution Approach 1:
The patent divides the prediction process into multiple stages: first obtaining intermediate prediction results for multiple targets simultaneously, then performing fusion processing on these intermediate results to obtain final target prediction results. This segmentation allows the system to maintain high productivity while improving accuracy through the fusion stage.
Solution Approach 2:
The patent performs preliminary prediction to obtain intermediate prediction results before final prediction. By pre-processing the prediction into intermediate results that can be fused together, the system prepares the necessary information in advance to achieve both efficiency and accuracy in the final prediction output.
2Measurement precision
If feature extraction and intermediate prediction are performed separately, then the prediction accuracy can be improved through fusion processing, but the training complexity increases
Solution Approach 1:
The patent combines the feature extraction network and prediction networks into a unified neural network structure where the feature extraction network shares parameters with the prediction networks. This merging allows the system to benefit from fusion processing for improved accuracy while reducing training complexity through parameter sharing and unified optimization.
Solution Approach 2:
The feature extraction network serves multiple functions: it extracts features for prediction and also serves as part of the prediction network itself through parameter sharing. This multi-functionality reduces the overall model complexity and training requirements while maintaining the benefits of fusion processing for improved prediction accuracy.
Data Source
AI summary
The present application relates to an object prediction method and apparatus, an electronic device, and a storage medium. The method is applied to a neural network and includes: performing feature extraction processing on a to-be-predicted object to obtain feature information of the to-be-predicted object; determining multiple intermediate prediction results for the to-be-predicted object according to the feature information; performing fusion processing on the multiple intermediate prediction results to obtain fusion information; and determining multiple target prediction results for the to-be-predicted object according to the fusion information. According to embodiments of the present application, feature information of a to-be-predicted object may be extracted; multiple intermediate prediction results for the to-be-predicted object are determined according to the feature information; fusion processing is performed on the multiple intermediate prediction results to obtain fusion information; and multiple target prediction results for the to-be-predicted object are determined according to the fusion information. The method facilitates improving the accuracy of multiple target prediction results.