Training Image Selection Using Structural Similarity Index
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning models used for image recognition in autonomous vehicles face performance degradation due to changes in objects during the conversion of simulation images to training images, leading to invalid training data that can affect model accuracy and safety.
Innovation Solution
An apparatus and method that detect the similarity between structures of objects in simulation and training images using a structural similarity index measure (SSIM), determining the validity of training images based on this similarity to prevent the use of degraded images for model training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If simulation images are converted to training images using image conversion devices, then the training data quantity is increased, but object changes occur during conversion leading to data validity degradation
Solution Approach 1:
The patent performs preliminary validation of training images by comparing simulation images with their converted training image counterparts before they are used for model training. The controller detects whether objects in training images correspond to objects in simulation images, and prevents degradation by filtering out invalid training images before they enter the training pipeline.
Solution Approach 2:
The patent introduces an intermediary validation mechanism that acts as a bridge between the image conversion process and the training process. This intermediary controller compares objects across simulation and training images to ensure correspondence, effectively mediating the quality control between data generation and model training stages.
2Productivity
If deep learning models are trained with converted simulation images, then training efficiency is improved, but model accuracy degrades due to object changes in converted images
Solution Approach 1:
The validation process is performed preliminarily before training, ensuring that only high-quality training images with corresponding objects are selected. This preliminary filtering maintains training efficiency by avoiding the need for post-training accuracy corrections while preventing accuracy degradation through proactive quality control.
Solution Approach 2:
The patent implements a feedback mechanism where the controller continuously monitors the correspondence between simulation and training images, providing feedback to filter out invalid training samples. This feedback loop ensures that only accurate training data enters the training process, maintaining both efficiency and model accuracy.
Data Source
AI summary
An apparatus for selecting a training image of a deep learning model and a method thereof are disclosed. The apparatus includes an input device and a controller. The input device receives a simulation image and information about an object in the simulation image from a simulation tool and receives a training image corresponding to the simulation image from an image conversion device. The controller detects a similarity between a structure of the object in the simulation image and a structure of an object in the training image and determines validity of the training image based on the detected similarity.