Feature Transformation Circuit for Low-Data Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous driving systems face challenges in accurately recognizing new objects with limited training data, as existing neural network technologies require substantial data for effective object detection.
Innovation Solution
An object detection device comprising a feature extracting circuit, a feature transforming circuit, and a decoder circuit that generates and transforms feature data using a transformation function, allowing for object detection regardless of object class, with the feature transforming circuit generating a transformation function using support feature data and reference data to produce transformed feature data less affected by object class, which is then decoded into a region map by the decoder circuit.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If a neural network is trained using a small number of data, then the data requirement is reduced, but the object detection performance deteriorates
Solution Approach 1:
The patent transforms feature data parameters by applying a transformation function that maps features from a source domain (with limited data) to a target domain. This parameter transformation allows the system to achieve good detection performance with small training data by leveraging features learned from source domain data that have been appropriately transformed rather than requiring extensive target domain training data.
Solution Approach 2:
The patent introduces a transformation function as an intermediary between source domain features and target domain detection tasks. This intermediary component learns the mapping relationship between different domains and enables knowledge transfer, allowing the system to overcome the limitation of small training data by using transformed features from a source domain as a bridge to achieve accurate detection in the target domain.
2Measurement precision
If a neural network is trained with extensive data to improve detection accuracy, then object detection performance improves, but the training complexity and resource requirements increase
Solution Approach 1:
By transforming feature data parameters through a learned transformation function, the patent enables the system to achieve high detection performance without requiring extensive training data. This parameter transformation approach simplifies the training process by reducing the amount of data needed while maintaining or improving detection accuracy, thereby reducing training complexity and resource requirements.
3Device complexity
If traditional object detection methods are used, then the system structure remains simple, but the ability to recognize new objects deteriorates
Solution Approach 1:
The patent applies parameter transformation to feature data, which enables the system to adapt to new objects by transforming source domain features to match target domain characteristics. This approach enhances the system's ability to recognize new objects without requiring complete retraining, as the transformation function can be applied to features from unseen object classes, improving adaptability while maintaining a relatively simple system structure.
Solution Approach 2:
The transformation function serves multiple purposes: it enables domain adaptation, facilitates new object recognition, and can be applied across different detection tasks. This universal component allows the system to handle various object types and domains with a single framework, enhancing versatility without proportionally increasing system complexity.
Data Source
AI summary
An object detecting device includes a feature extracting circuit configured to extract first feature data from an input image; a feature transforming circuit configured to transform the first feature data into transformed feature data according to a transformation function; and a decoder circuit configured to decode the transformed feature data into a region map indicating a detected object.