Object Detection Dataset Construction via Image Entropy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Constructing a high-quality dataset for deep learning models is costly and time-consuming, requiring manual inspection and often results in missing or incorrectly created object information, which hampers the performance of deep learning models.
Innovation Solution
A data processing device that uses image entropy to prioritize and construct an object detection dataset by computing image entropy through image segmentation algorithms or object detection models, identifying and re-processing images with high entropy values to ensure accurate data creation, thereby reducing costs and improving efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual inspection and construction methods are used to create high-quality datasets, then data accuracy is improved, but construction time and costs increase significantly
Solution Approach 1:
The system performs self-inspection by automatically computing image entropy and identifying images that require re-processing. The dataset construction system inspects its own output quality without manual intervention, using entropy computation to detect missing or incorrect object information automatically.
Solution Approach 2:
Manual inspection processes are replaced with automated computational methods. Image entropy computation algorithms substitute for human inspectors, automatically identifying images with quality issues based on entropy thresholds and triggering re-processing workflows.
2Manufacturing precision
If comprehensive manual inspection is performed on all images, then data quality is improved, but resource utilization and costs increase
Solution Approach 1:
Instead of uniformly inspecting all images, the system applies quality inspection selectively based on local characteristics. Image entropy computation identifies specific images that deviate from quality standards, allowing targeted re-processing only of problematic images rather than comprehensive inspection of the entire dataset.
Solution Approach 2:
The system uses image entropy as a quality parameter to dynamically determine which images require inspection. By computing entropy values and comparing them against thresholds, the system adapts its inspection strategy based on the inherent quality characteristics of each image, optimizing resource allocation.
3Productivity
If all images are processed equally without prioritization, then processing completeness is maintained, but processing efficiency decreases
Solution Approach 1:
The system performs preliminary quality assessment by computing image entropy before final dataset construction. This preliminary action identifies high-entropy images that are likely to contain quality issues, allowing prioritized processing and inspection of these images while maintaining overall processing completeness.
Solution Approach 2:
The system implements a feedback loop where image entropy computation results inform subsequent processing decisions. Images exceeding entropy thresholds are flagged for re-processing, and the system continuously monitors quality metrics to adjust processing priorities, ensuring both efficiency and completeness.
Data Source
AI summary
Disclosed are an object detection dataset construction method using image entropy and a data processing device performing the same. The data processing device includes an input unit configured to receive multiple images, a control unit configured to choose processing priorities of the received multiple images using image entropy and construct an object detection dataset from a corresponding image according to the chosen processing priorities, and a storage unit configured to store the constructed object detection dataset.