Object Detection Learning With Size-Aware Ground Truth Targets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection systems face challenges in generating accurate ground truth targets due to varying location dependency of IoU based on the size of the ground truth area, which affects the learning process and detection accuracy.
Innovation Solution
A learning system that generates ground truth targets by considering the size of the ground truth area, using a score indicating the degree of area matching between area candidates and the ground truth area, and adjusting thresholds or scores accordingly to improve target generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If IoU is used as the sole criterion for generating ground truth targets, then the generation process is simple, but the detection accuracy deteriorates due to location dependency varying with ground truth area size
Solution Approach 1:
The patent applies local quality by making the ground truth target generation process adaptive to different ground truth area sizes. Specifically, the area candidate selection criteria and scoring mechanisms are adjusted based on whether the ground truth area is small or large, allowing the system to optimize for location dependency characteristics specific to each size category rather than using a uniform approach
Solution Approach 2:
The patent changes parameters related to area candidate selection by introducing size-dependent thresholds and scoring weights. The system modifies the IoU threshold values and selection criteria based on the ground truth area size, transforming the generation process from a fixed-parameter approach to a dynamic parameter adjustment approach that adapts to different detection scenarios
2Productivity
If the ground truth target generation does not consider ground truth area size, then the generation process is fast, but the learning effectiveness deteriorates due to improper target generation for different sized objects
Solution Approach 1:
The patent segments the ground truth target generation process into distinct pathways based on ground truth area size. It divides objects into small and large categories and applies different generation strategies to each segment, allowing efficient processing within each segment while maintaining overall effectiveness through specialized handling of size-specific characteristics
Solution Approach 2:
The patent introduces dynamic adjustment mechanisms that automatically adapt the generation process based on the detected ground truth area size. The system dynamically selects appropriate thresholds, scoring criteria, and selection strategies based on real-time size assessment, enabling the generation process to be both efficient and effective across varying object sizes
Data Source
AI summary
A learning system includes: an area candidate generation unit that generates an area candidate that is a candidate for an area in which an object is detected from an image; a ground truth area acquisition unit that obtains information about a ground truth area including the object; and a ground truth target generation unit that generates a ground truth target used for learning of an object detection unit that detects the object from the image, on the basis of a score indicating a degree of area matching between the area candidate and the ground truth area and a size of the ground truth area. According to such a learning system, it is possible to properly perform the learning in accordance with the size of the ground truth area.