Learning Data Selection Using Metadata Similarity Across Scenes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in efficiently adding data related to similar scenes different from the target scene for machine learning, particularly in scenarios where data collection is time-consuming, difficult, or restricted, affecting recognition model accuracy.
Innovation Solution
A method for generating learning data involves acquiring and analyzing first and second data, generating metadata for both, and determining similarity to select third data for machine learning based on metadata comparison, enabling efficient addition of similar scene data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large amount of data of various scenes similar to the target scene is collected to improve recognition accuracy, then the recognition accuracy is improved, but the time required for data collection increases significantly
Solution Approach 1:
The patent creates virtual copies of target scenes through 3D modeling and rendering. Instead of collecting real-world data for every possible scene variation, the system generates synthetic images by copying and transforming 3D models, lighting conditions, and camera parameters to simulate diverse scenes efficiently without time-consuming field collection
Solution Approach 2:
The patent performs preliminary 3D modeling and scene reconstruction before data collection. By pre-building accurate 3D models of target scenes and pre-calculating various viewing angles and lighting conditions, the system prepares all necessary scene variations in advance, eliminating the need for time-consuming on-site data collection for each scene type
2Measurement precision
If data of various scenes similar to the target scene is collected to improve recognition accuracy, then the recognition accuracy is improved, but the difficulty and complexity of data collection increases
Solution Approach 1:
The patent replaces complex mechanical data collection systems with computational methods. Instead of using multiple cameras, sensors, and physical scene setups to capture diverse scenes, the system uses 3D modeling software and rendering algorithms to generate synthetic scene data, dramatically simplifying the data collection process while maintaining scene diversity
Solution Approach 2:
The patent creates a universal 3D modeling framework that can generate data for multiple scene types and variations from a single set of 3D models. This multi-functional system can produce diverse training data by varying parameters such as lighting, camera angles, and object positions without requiring separate collection systems for each scene type
Data Source
AI summary
First data that is at least one of data used for machine learning of a recognition model that recognizes a feature related to a predetermined scene and data erroneously recognized by the recognition model is acquired, first metadata indicating a characteristic of the first data is generated, second data related to a scene different from the predetermined scene is acquired, second metadata indicating a characteristic of the second data is generated, whether to newly select the second data as third data to be used for the machine learning based on a similarity between the first metadata and the second metadata is determined, and the third data is output.