Object Detection Network Training via Clustered Prior Box Parameters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face detection methods in object detectors struggle with accurately generating positive samples during training due to manually set parameters for prior boxes, leading to insufficient learning and reduced detection accuracy.
Innovation Solution
The method involves an image processing technique where an object detection network is trained using cluster center values obtained by clustering labeled regions in a sample image set, allowing for iterative training and improved overlap between predicted and labeled bounding boxes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If manual parameter setting for prior boxes is used, then device complexity is reduced, but detection accuracy deteriorates due to insufficient positive samples
Solution Approach 1:
The system performs self-service by automatically generating prior box parameters through clustering algorithms based on the training data distribution, eliminating the need for manual parameter setting while ensuring optimal detection accuracy through data-driven parameter selection
Solution Approach 2:
The invention dynamically changes prior box parameters (area, aspect ratio) by computing cluster center values from the training data, allowing the parameters to adapt to the specific characteristics of the target objects in the dataset rather than using fixed manual settings
2Ease of manufacture
If fixed prior box parameters are used, then training process is simplified, but positive sample generation is insufficient leading to reduced learning effectiveness
Solution Approach 1:
The system performs preliminary action by pre-computing cluster center values from the training data before actual detection, generating a comprehensive set of prior box parameters that cover all possible target configurations, thereby ensuring sufficient positive samples are available during training without complicating the training process itself
3Adaptability or versatility
If manual parameter setting is used, then adaptability to different datasets is reduced, but parameter adjustment becomes more controllable
Solution Approach 1:
The invention introduces dynamics by making prior box parameters adaptive rather than static, allowing the system to automatically adjust parameters based on the specific characteristics of different datasets through clustering, while maintaining ease of operation through automated computation without requiring manual intervention for each new dataset
Data Source
AI summary
An image processing method and device, an equipment, and a computer-readable storage medium are provided. The method includes the following. An image is obtained. Object detection is performed on the image with an object detection network to obtain an object detection result. The object detection network is a network obtained by performing object detection on sample images in a sample image set based on at least one cluster center value with an initial object detection network and training iteratively the initial object detection network based on results of the object detection. The at least one cluster center value is obtained by clustering labeled regions in a labeled region set corresponding to the sample image set based on preset attribute information. The image is processed based on the object detection result.