Style-Coded Data Generation Framework for Realistic 3D Outputs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional simulators struggle to generate realistic data in virtual scenarios, especially for designing three-dimensional objects, and are limited in applying customized conditions.
Innovation Solution
A data generation framework utilizing a processor and storage system that includes a style encoder, mapping network, generator, and discriminator to learn and judge data quality, enabling high-quality, realistic data generation through a series of encoding, noise addition, prediction, and discrimination processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a conventional simulator is used to generate images according to customized conditions, then the generation process can be controlled with specific parameters (such as weather, time, spatial conditions), but the generated images are not realistic and cannot be effectively applied in designing three-dimensional objects
Solution Approach 1:
The patent introduces an image translation model as an intermediary between the conventional simulator and the final output. This model takes the generated images and transforms them into more realistic versions while preserving the customized conditions, thereby improving image realism without losing the controllable parameters needed for 3D object design applications
Solution Approach 2:
The patent employs parameter changes by adjusting various attributes of the generated images through the image translation model. By modifying parameters such as texture, lighting, and material properties while maintaining the original customized conditions (weather, time, spatial conditions), the system enhances image realism and applicability for 3D object design
2Manufacturing precision
If conventional simulation methods are used, then the process is simple and controllable, but the quality and authenticity of the generated data are insufficient
Solution Approach 1:
The patent merges the conventional simulator with an image translation model into a unified data generation system. This combination allows the system to maintain the simplicity and controllability of the simulator while incorporating the enhanced data quality capabilities of the translation model, achieving high-quality data generation without excessive complexity
Solution Approach 2:
The patent uses the image translation model to create improved copies of the images generated by the conventional simulator. Instead of replacing the simple simulator, the system generates initial images and then creates enhanced versions through translation, thereby improving data quality while maintaining the simplicity of the original generation process
Data Source
AI summary
A method and an apparatus related to a data generation framework are provided. A first code is obtained from reference data by a style encoder. A second code is obtained from latent encoding by a mapping network. The first code and the second code are both a style code, and the style code corresponds to one or more style options. First source data is input to a generator, and a first output corresponding to the first source data is output by referring to the style code by the generator. The first output is input to a discriminator, and a second output corresponding to the first output is output by the discriminator. The second output represents whether the first output corresponds to the style option.