Neural Network Training Data Encoding for Cross-Dataset Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural networks trained on multiple datasets face reduced performance due to misclassification of negative samples as positive samples across different datasets, leading to inaccurate object detection and classification.
Innovation Solution
Introduce individual background classes for each dataset and assign ignore attributes to object classes and background classes that are unique to other datasets, ensuring that these classes do not contribute to the neural network's training process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If multiple datasets are used for training neural networks, then the quantity of training data increases, but the detection accuracy decreases due to misclassification of negative samples as positive samples
Solution Approach 1:
The patent divides the background class into multiple dataset-specific background classes (e.g., background_class_1, background_class_2) corresponding to different datasets. This segmentation prevents samples from being misclassified across datasets by assigning each sample to the appropriate background class based on its source dataset, thereby maintaining detection accuracy while using multiple datasets for training.
Solution Approach 2:
The patent introduces an ignore attribute as an intermediary mechanism that identifies and excludes problematic samples from training. This attribute marks samples that should not be used for training the neural network, preventing misclassification issues while still allowing the benefits of multiple datasets to be utilized.
2Ease of manufacture
If traditional annotation methods are used for multiple datasets, then the annotation process is simple, but the workload and time for preparation increases significantly
Solution Approach 1:
The patent performs preliminary encoding of training data before the actual training process. By pre-processing and encoding the data structure with dataset-specific background classes and ignore attributes in advance, the system reduces the time and effort required during the training phase, making the overall process more efficient.
Solution Approach 2:
The patent changes the parameter structure of the training data by introducing dataset-specific background class identifiers and ignore attributes. This parameter transformation allows the system to handle multiple datasets more efficiently, reducing annotation preparation time while maintaining ease of manufacture.
3Adaptability or versatility
If all object classes are annotated in every dataset, then the training data is comprehensive, but the device complexity increases
Solution Approach 1:
The patent extracts and removes unnecessary annotations from the training data structure. By using the ignore attribute to identify samples that should not be trained on, the system can selectively exclude redundant annotations while maintaining comprehensive coverage of important object classes, thereby reducing structural complexity.
Solution Approach 2:
The patent applies partial annotation by requiring object classes to be annotated only in datasets where they are relevant, rather than mandating annotations across all datasets. This partial action approach maintains adaptability and versatility while significantly reducing the complexity of the overall data structure.
Data Source
AI summary
A method for encoding training data for training of a neural network comprises: obtaining training data including multiple datasets, each dataset comprises images annotated with at least one respective object class, forming, each dataset having an individual background class associated with the object class; encoding the images of the datasets to be associated with their respective individual background class; encoding image patches belonging to annotated object classes to be associated with their respective object class; encoding each of the datasets, to include an ignore attribute (“ignore”) to object classes that are annotated only in the other datasets and to background classes formed for the other datasets of the multiple datasets, the ignore attribute indicating that the assigned object class and background classes do not contribute in adapting the neural network in training using the respective dataset; and providing the encoded training data for training of a neural network.