Training Image Generation System with Adjustable Recognition Difficulty
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating training images for object recognition, such as AI or contraband inspection, face challenges in ensuring uniform quality and adjustable recognition difficulty, leading to poor performance when sample differences are too great or too small, and are not customizable to user demands.
Innovation Solution
An automatic generation system that acquires container images, selects target images, and repeatedly adds them to the container images until a reliability threshold for recognition difficulty is met, allowing for adjustable recognition difficulty through user-modifiable parameters, ensuring consistent and customizable training images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If training images are generated using machine learning or random number methods, then a large number of samples can be produced, but the quality of samples is not uniform and recognition difficulty cannot be adjusted
Solution Approach 1:
The patent changes the parameters of container images (such as image type, size, quantity, arrangement, and transparency) to control the recognition difficulty of training samples. By adjusting these parameters, the system generates samples with uniform quality and controllable difficulty levels, resolving the contradiction between producing large quantities of samples and maintaining their quality uniformity.
2Productivity
If only low difficulty samples are used for training, then training can be completed quickly, but the model cannot recognize high difficulty samples
Solution Approach 1:
The patent introduces a dynamic difficulty adjustment mechanism where the recognition difficulty of training samples can be flexibly changed based on training progress and requirements. The system can switch between generating easy and difficult samples by adjusting container image parameters, allowing the model to be trained efficiently while maintaining the capability to recognize samples at various difficulty levels.
3Adaptability or versatility
If samples with huge differences are used for training, then diverse recognition scenarios can be covered, but the training performance deteriorates due to non-uniform quality
Solution Approach 1:
The patent applies local quality control by systematically varying specific parameters of container images (image type, size, quantity, arrangement, transparency) while maintaining overall sample quality uniformity. This allows the generation of diverse training scenarios with controlled differences, ensuring both coverage of recognition scenarios and maintenance of training performance.
Data Source
AI summary
An automatic generation system of a training image and a method thereof are provided. The disclosure generates a training image and records the target category and the target position. The disclosure adds the target image to the container image as a candidate image, calculates a reliability of the candidate image, and repeatedly executes the process until the reliability of the candidate image meets a threshold condition for generating the training image. The disclosure is able to generate the training images automatically, and the recognition difficulty of the training image is adjustable by the user, so as to be suitable for customized recognition training.