Bright Spot Image Generation Training for Small Region Depiction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to visually depict or represent small regions and anomalous regions in generated images, such as leakage regions in medical images, in a plausible manner.
Innovation Solution
A learning system and method for training an image generation model that includes acquiring training data with input and correct answer images, generating images based on these data, and updating the model to improve the depiction of bright spot regions by adjusting the foreground-to-background ratio and using techniques like U-Net and generative adversarial networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image generation methods are used, then overall image generation is achieved, but small regions and anomalous regions cannot be visualized or depicted in a plausible manner
Solution Approach 1:
The patent divides the image generation task into two segments: a global image generation model for overall structure and a local bright spot image generation model for small regions and anomalous areas. This segmentation allows each model to specialize, with the local model specifically trained to accurately visualize small regions that would otherwise be indistinguishable in the global image.
Solution Approach 2:
The patent applies local quality by creating a dedicated bright spot image generation model that focuses specifically on local regions with abnormal brightness. This model is trained separately to enhance the depiction quality of small and anomalous regions, ensuring they are visualized with appropriate detail and plausibility rather than being lost in the global image context.
2Shape
If the image generation model is trained on full images, then overall structure is preserved, but details of small regions are lost
Solution Approach 1:
The training process is segmented into two stages: first training the global image generation model on full images to preserve overall structure, then training the local bright spot image generation model on extracted bright spot regions to capture fine details. This segmented training approach ensures both global structure and local details are optimized.
Solution Approach 2:
The patent extracts bright spot regions from training images to create specialized training data for the local model. By taking out these small regions with abnormal brightness and training a dedicated model on them, the system captures detailed characteristics that would be imperceptible in full-image training, thereby improving detail accuracy.
Data Source
AI summary
A learning system, for performing training of an image generation model configured to output a generated image having one or more bright spot regions and corresponding to an input image, acquires training data including an input image and a correct answer image having one or more bright spot regions and corresponding to the input image; inputs the input image to the image generation model to acquire a generated image; acquires, based on the correct answer image, a first bright spot image including at least one bright spot region included in the one or more bright spot regions; acquires, based on the generated image obtained by inputting the input image to the image generation model, a second bright spot image corresponding to the first bright spot image; and updates the image generation model based on an error between the first bright spot image and the second bright spot image.