Object Detection Model Iterative Negative Sample Enhancement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection models fail to consider the influence of negative samples during training, leading to increased false detections due to the loss of background information, which affects detection accuracy.
Innovation Solution
An object detection model generation method that iteratively determines detection accuracy at each training node, enhances mis-detected negative samples, and trains the model using these enhanced samples to improve false detection rates, incorporating a preset amount of training samples and negative sample enhancement rules.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If random cropping and random matting are performed according to labeled box position and size to obtain positive samples, then training data diversity is improved, but background information is lost leading to increased false detections
Solution Approach 1:
The patent applies preliminary action by pre-processing negative samples through background extraction before they are used in training. The background extraction module extracts background images from original images by removing foreground objects, and this pre-processed background information is then used to generate negative samples. This preliminary preparation ensures that background information is preserved and properly utilized during the training process, preventing the loss of crucial contextual information that would otherwise lead to false detections.
Solution Approach 2:
The patent converts the harmful effect of lost background information into a benefit by systematically extracting and utilizing this background information to generate enhanced negative samples. Instead of treating background loss as a negative consequence of random cropping, the invention transforms it into a useful resource by creating dedicated background samples that are specifically designed to improve the model's ability to distinguish between actual objects and background regions, thereby reducing false detections.
2Ease of operation
If existing training methods are used that focus on positive samples, then training process simplicity is maintained, but false detections increase due to lack of negative sample consideration
Solution Approach 1:
The patent applies segmentation by dividing the training process into distinct modules: a background extraction module that separates background from foreground, a negative sample generation module that creates training samples from extracted backgrounds, and a model training module that uses these samples. This segmentation allows the complex task of handling negative samples to be broken down into manageable, independent steps, maintaining ease of operation while significantly improving detection accuracy through systematic negative sample processing.
Data Source
AI summary
An object detection model generation method as well as an electronic device and a computer readable storage medium using the same are provided. The method includes: during the iterative training of the to-be-trained object detection model, the detection accuracy of the iteration nodes of the object detection model is sequentially determined according to the node order, and the mis-detected negative samples of the object detection model at the iteration nodes with the detection accuracy less than or equal to a preset threshold are enhanced. Then the object detection model is trained at the iteration node based on the enhanced negative samples and a first amount of preset training samples. After the training at the iteration nodes are completed, it returns to the step of sequentially determining the detection accuracy of the iteration nodes of the object detection model until the training of the object detection model is completed.