Distortion Function Enhances ROI Resolution in Generative AI Synthetic Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Generating high-resolution synthetic data remains a challenge due to uneven distribution of data elements in source data, leading to overfitting and under-learning of crucial features, especially in regions of interest like the mouth in digital human images, resulting in anomalous artifacts and poor rendering of small objects.
Innovation Solution
A method that distorts source data using a distortion function to enhance regions of interest (ROIs) relative to other regions, generating a distorted ground truth that guides a generative AI model to focus on sparse data areas, such as the mouth, thereby improving resolution without increasing computational burden.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If source data is used directly for training generative AI models, then the training process is simple and fast, but the resolution of regions of interest (ROIs) in synthetic data is poor and artifacts appear
Solution Approach 1:
The patent applies preliminary action by distorting the source data before feeding it to the generative AI model. Specifically, regions of interest (ROIs) are identified and distorted to occupy a larger portion of the training data, which prepares the data in advance to ensure the model learns high-resolution features of ROIs. This preprocessing step resolves the contradiction by improving ROI resolution without requiring complex modifications to the model architecture itself.
Solution Approach 2:
The patent implements local quality by applying different distortion levels to different regions of the source data. ROIs are selectively distorted to occupy enhanced portions of the training data, while non-ROI regions maintain their original characteristics. This allows the system to focus computational resources on improving the quality of critical regions without unnecessarily complicating the processing of all data regions.
2Manufacturing precision
If source data is distorted to enhance ROIs, then the resolution of synthetic ROIs improves, but the training process becomes more complex
Solution Approach 1:
The patent applies parameter changes by modifying the spatial distribution parameters of source data through distortion. The distortion function changes how ROIs are represented in the training data, making them occupy a larger effective area. This parameter transformation allows the model to learn high-resolution ROI features more efficiently, potentially reducing the number of training iterations needed despite the added distortion step.
Solution Approach 2:
By pre-distorting the source data to enhance ROI representation, the patent performs the complex transformation work before training begins. This preliminary action ensures that the model receives optimally structured data from the start, which can accelerate convergence during training compared to attempting to achieve the same effect through iterative model adjustments.
3Manufacturing precision
If traditional synthetic data generation is used, then computational resources are saved, but the synthetic data contains artifacts and poor rendering of small objects
Solution Approach 1:
The patent applies local quality by selectively enhancing only the regions of interest in the training data, rather than uniformly processing all data. This allows the system to concentrate computational resources on improving the rendering of small objects and ROIs, achieving better quality where it matters most without proportionally increasing overall computational burden.
Solution Approach 2:
The distortion function changes the spatial parameters of ROIs in the training data, effectively magnifying their representation. This parameter transformation allows the model to learn fine details of small objects with the same computational resources, improving rendering quality without requiring additional hardware or exponentially more training data.
Data Source
AI summary
Synthesizing high-resolution data includes distorting, with a distortion function, a region of interest (ROI) within an input of inferential data. The distorting generates distortion data within which the ROI is enhanced relative to other regions of the distortion data. A generative artificial intelligence (AI) model generates synthetic data in response to input of the distortion data. The generative AI model is trained against a distorted ground truth generated using the distortion function to distort one or more regions of interest ROI within source data used to guide the generative AI model in generating the synthetic data.