Random Graphic Dataset Generation for Hologram Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image training datasets, such as ImageNet, CIFAR, AFLW, and INRIA Person Dataset, are limited in data, leading to poor performance of neural network models, particularly in generating holograms.
Innovation Solution
A method and system for generating a random graphic training dataset by randomly generating graphic datasets containing squares, circles, and equilateral triangles, performing gradual filling, and applying translation transformations to create a comprehensive training dataset.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If traditional image training datasets (ImageNet, CIFAR, AFLW, INRIA Person Dataset) are used, then the datasets are available and easy to obtain, but the data quantity is limited and cannot provide all possible training data, resulting in poor hologram generation effect
Solution Approach 1:
The patent applies preliminary action by pre-defining graphic templates (squares, circles, triangles) with predetermined geometric properties before actual training data generation. These templates serve as foundational elements that are subsequently transformed through various operations to generate diverse training data, enabling comprehensive coverage of possible training scenarios without requiring exhaustive collection of real-world images
Solution Approach 2:
The patent employs parameter changes by systematically varying geometric parameters (position, size, rotation angle, transparency) of basic graphic templates to generate diverse training samples. This approach transforms a limited set of template definitions into a vast combinatorial space of training data, effectively increasing data quantity while maintaining controlled quality through parameter-based generation rather than random collection
2Reliability
If more comprehensive training data is provided to improve neural network training effect, then the training effect improves, but the complexity of data generation increases
Solution Approach 1:
The patent applies segmentation by decomposing the complex task of generating comprehensive training data into independent, manageable components: basic graphic templates (squares, circles, triangles), transformation operations (translation, rotation, scaling), and composition rules. This modular approach allows each component to be independently defined and combined, reducing overall system complexity while achieving comprehensive training data coverage
Solution Approach 2:
The patent implements universality by designing basic graphic templates that serve multiple functions simultaneously. The same template set (squares, circles, triangles) is used across different training scenarios and can be transformed through various operations to generate diverse samples. This multi-functional approach eliminates the need for separate data generation systems for different training needs, reducing complexity while maintaining comprehensiveness
Data Source
AI summary
A method and system for generating a random graphic training dataset and a hologram is disclosed, which relates to the technical field of image processing. The method for generating a random graphic training dataset includes: randomly generating a plurality of graphic datasets, where each of the graphic datasets includes a square, a circle and an equilateral triangle; for any graphic dataset, performing gradual filling on each graphic in the graphic dataset to obtain a gradual graphic dataset; and performing translation transformation on each graphic in the gradual graphic dataset for a predetermined number of times to obtain random graphics corresponding to the graphic dataset, and determining random graphics corresponding to all graphic datasets as a random graphic training dataset. More comprehensive training data can be provided and a training effect of a model can be improved.