Neighbor Bounding Box Aggregation for Complete Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Generating bounding box information in neural networks requires significant memory and computing resources, and existing methods may miss parts of objects due to selecting bounding boxes with maximum confidence that do not fully surround them.
Innovation Solution
A system that aggregates bounding box proposals by selecting strong neighbors based on intersection-over-union (IoU) values to update confidence and coordinates, using techniques like non-maximum suppression and neural network models to enhance bounding box determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If bounding box proposals are generated using neural networks, then object detection capability is improved, but memory and computing resources are significantly consumed
Solution Approach 1:
The patent extracts and processes only the most relevant bounding box proposals by filtering out redundant ones. It selects a subset of proposals that are most likely to contain the object based on confidence scores and spatial relationships, thereby reducing the computational load while maintaining detection accuracy.
Solution Approach 2:
The patent segments the bounding box proposals into different categories based on their confidence scores and spatial overlaps. By dividing the set of proposals into meaningful groups, the system can process them more efficiently using targeted computational strategies rather than treating all proposals uniformly.
2Speed
If maximum confidence bounding box is selected, then processing speed is improved, but object completeness is reduced due to missed parts
Solution Approach 1:
The patent merges multiple bounding box proposals into a single comprehensive bounding box by combining information from several proposals. This approach integrates the strengths of multiple detections to produce a complete and accurate bounding box that encompasses the entire object, overcoming the limitations of selecting only the maximum confidence box.
Solution Approach 2:
The patent uses feedback mechanisms to iteratively refine the bounding box selection process. By evaluating the overlap and confidence scores of multiple proposals and adjusting the selection based on this feedback, the system achieves both speed and completeness in bounding box generation.
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.