Anchor-Based Training Data for Efficient Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning-based object detection systems face a trade-off between increasing accuracy and processing efficiency due to the use of multiple anchor shapes, which leads to higher computational demands.
Innovation Solution
A data creation method that associates area information with image data using anchors to create training data, allowing for high-accuracy object detection while minimizing processing requirements, involving a training data creation device and a program creation device that perform deep learning to generate a trained program for object detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the number of anchors increases to improve detection accuracy, then detection accuracy is improved, but the amount of arithmetic processing increases
Solution Approach 1:
The patent performs preliminary action by determining frame information based on anchor positions during the training data creation phase. This preprocessing step organizes the relationship between anchors and ground truth data before deep learning training, allowing the model to learn more efficiently from structured data where each anchor's corresponding frame is pre-determined, thus reducing the computational burden during actual detection while maintaining high accuracy
Solution Approach 2:
The patent segments the detection task by associating each anchor with specific frame information that divides the image into cells. This segmentation approach allows the deep learning model to process detection tasks in a structured manner where anchors are evaluated against specific frame regions, enabling efficient computation even with multiple anchor types by organizing the processing into discrete cell-based evaluations
2Measurement precision
If the choice of anchor to be used increases to improve detection accuracy, then detection accuracy is improved, but the amount of processing when deciding conditions for deep learning increases
Solution Approach 1:
The system performs preliminary action by pre-determining frame information for each anchor during training data preparation. This includes calculating which frames (cells) each anchor should be associated with before the deep learning training begins. By organizing this anchor-frame relationship in advance, the system reduces the time needed for condition decision-making during training, as the model receives pre-structured data indicating which anchors correspond to which ground truth frames
Solution Approach 2:
The patent introduces frame information as an intermediary element that mediates between anchors and ground truth data. This intermediary structure organizes the complex relationships between multiple anchor types and various objects by assigning each anchor to specific frame regions. This intermediary layer simplifies the training process by providing a clear mapping structure, reducing the time required for the system to decide training conditions even when dealing with diverse anchor choices
Data Source
AI summary
A data creation device that is a training data creation device for creating training data which is used for deep learning of an object detection program for detecting whether or not an object is included in an image, includes: an acquisition unit that acquires an anchor that is information on a frame specifying an area for each cell for detecting presence or absence of the object from the image; and a creation unit that associates area information of the object with image data to create the training data that includes a plurality of image data in which the area information is included. The creation unit determines a frame of the area information, based on a position of the anchor.