Bounding Box Aggregation Using IoU Neighbors for Full Object Coverage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Generating bounding box information is resource-intensive in terms of memory and computing, and existing methods may miss parts of objects due to selecting maximum confidence bounding boxes that do not fully encompass them.
Innovation Solution
A system that aggregates bounding box proposals by identifying strong neighbors with intersection-over-union (IoU) values to update confidence and coordinates, using techniques like non-maximum suppression and neural network models to refine bounding box determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If maximum confidence bounding boxes are selected, then processing speed is improved, but object coverage completeness deteriorates
Solution Approach 1:
The patent merges multiple bounding box proposals with high confidence scores into a single refined bounding box by aggregating their coordinate information. This combining approach maintains processing efficiency while improving object coverage completeness, as the merged bounding box encompasses the entire object better than any single proposal box.
Solution Approach 2:
The patent performs preliminary sorting of bounding box proposals by confidence score before aggregation. This preliminary action ensures that the most reliable proposals are processed first and given higher weight in the aggregation, maintaining speed while improving completeness through selective processing.
2Measurement precision
If multiple bounding box proposals are processed, then accuracy is improved, but computing resources consumed increase
Solution Approach 1:
The patent processes only the top K bounding box proposals with highest confidence scores rather than all proposals. This partial action approach improves accuracy by considering multiple proposals while reducing computing resource consumption by limiting the number of proposals processed to a manageable subset.
Solution Approach 2:
The patent changes the parameter of confidence score threshold dynamically. By adjusting which proposals are selected for aggregation based on their confidence scores, the system optimizes the balance between accuracy and computing resource consumption, processing only the most promising proposals.
3Reliability
If bounding box aggregation is performed, then object coverage completeness is improved, but processing time increases
Solution Approach 1:
The patent segments the bounding box aggregation process into distinct stages: sorting proposals by confidence, selecting top K proposals, and aggregating their coordinates. This segmentation allows each stage to be optimized independently, improving object coverage completeness while controlling processing time through efficient stage execution.
Solution Approach 2:
The patent creates a simplified representation of bounding box proposals for aggregation, copying only the essential coordinate and confidence information needed for the aggregation process. This copying approach reduces the complexity of processing while maintaining object coverage completeness.
Data Source
AI summary
Apparatuses, systems, and techniques to generate bounding box information. In at least one embodiment, for example, bounding box information is generated based, at least in part, on a plurality of candidate bounding box information.