Twin Network Joint Training for Image Recognition Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image recognition technologies face low accuracy due to separate training methods for recognition and retrieval tasks, which do not consider the correlation between the two, leading to poor performance in fine-grained recognition and retrieval.
Innovation Solution
Joint training of twin networks with consistent structures and shared weights for both recognition and retrieval processing within a classification network, using a shared set of training data to enhance feature extraction and classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate training methods are used for recognition and retrieval tasks, then the training process is simple and independent, but the accuracy of image recognition is low due to not considering the correlation between tasks
Solution Approach 1:
The patent merges the recognition and retrieval training processes into a single joint training framework. The classification network simultaneously trains both the first sub-network (recognition) and second sub-network (retrieval) using combined training data, allowing the models to learn correlated features and relationships between the two tasks, thereby improving recognition accuracy while managing training complexity through unified architecture design
2Measurement precision
If joint training of twin networks is implemented, then feature extraction accuracy is improved, but the training time and computational resources increase
Solution Approach 1:
The patent implements universal twin networks where the first and second sub-networks share common structural components and training mechanisms. Both networks use the same backbone architecture and training framework, allowing them to benefit from shared computational resources and learned representations. This multi-functionality approach improves feature extraction accuracy for both recognition and retrieval tasks while reducing redundant computational overhead compared to training completely separate models
3Productivity
If separate training is used for recognition and retrieval, then the training process is efficient and fast, but the correlation between recognition and retrieval tasks is not captured
Solution Approach 1:
The joint training process incorporates feedback mechanisms where the performance and learned representations from both recognition and retrieval tasks continuously refine the shared features and parameters. The training system monitors and adjusts based on the interplay between the two tasks, allowing the model to capture correlations and dependencies that would be missed in separate training, thereby improving reliability while maintaining training efficiency through integrated optimization
Data Source
AI summary
An image recognition method, electronic device, and storage medium are provided and relate to the fields of artificial intelligence, computer vision, deep learning, image processing, and the like. The method includes: performing joint training on a first sub-network configured for recognition processing and a second sub-network configured for retrieval processing in a classification network by adopting an identical set of training data to obtain a trained target classification network, wherein, the first sub-network and the second sub-network are twin networks that are consistent in network structures and share a set of weights; and inputting image data to be recognized into the target classification network to obtain a recognition result. By adopting the method, the accuracy of the image recognition may be improved.