Learning Data Generation via Positive to Negative Instance Conversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection systems face challenges in accurately distinguishing between intended subjects and similar objects, leading to frequent erroneous detections, especially when objects resembling subjects are present in images.
Innovation Solution
A learning data generating apparatus that converts positive instance teacher data into negative instances based on preset rules, using a data acquisition unit and generation unit to create generated learning data, which includes applying specific image conversion conditions to reduce erroneous detections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing learning data is expanded to multiple positive instance data, then the robustness against background changes is improved, but the ability to restrain erroneous detections of non-subjects deteriorates
Solution Approach 1:
The patent inverts the conventional approach by converting positive instance teacher data into negativeinstance teacher data. Instead of expanding existing positive data, the system creates negative examples by applying conversion rules to positive data, thereby teaching the detection model what subjects are not. This inversion enables the model to distinguish between subjects and non-subjects more effectively, resolving the contradiction between robustness and detection accuracy.
Solution Approach 2:
The patent changes the parameter of data labeling by converting positive labels to negative labels through systematic transformation rules. By modifying the label parameter and applying conversion rules (such as adding conversion data to learning data), the system creates negative instances that improve the model's ability to restrain erroneous detections while maintaining robustness against background changes.
2Measurement precision
If many images of non-subjects are used to restrain erroneous detections, then the accuracy of subject detection is improved, but the complexity of data acquisition and processing increases
Solution Approach 1:
The patent creates negative instance data by copying and transforming existing positive instance data through conversion rules, rather than acquiring completely new non-subject images. This copying approach (such as creating synthetic negative examples by modifying positive data) significantly reduces the complexity of data acquisition while still providing effective negative examples for training the detection model to distinguish subjects from non-subjects.
Solution Approach 2:
The system uses itself to generate the data it needs. By applying conversion rules to existing positive teacher data, the learning data generating apparatus creates negative instance data autonomously without requiring external data acquisition. This self-service mechanism simplifies the overall system complexity while maintaining high detection accuracy through effective negative example generation.
Data Source
AI summary
In order to provide a learning data generating apparatus that is able to efficiently restrain erroneous detections, the learning data generating apparatus includes a data acquisition unit configured to acquire learning data including teacher data, and a generation unit configured to generate generated learning data based on the learning data and a generating condition, wherein the generation unit converts teacher data of a positive instance into teacher data of a negative instance according to a preset rule when generating the generated learning data.