Image File Generating Device for Standardized Training Data

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Generating an inference model that aligns with a user's specific intentions and concerns is complicated due to the lack of standardized and simplified methods for creating training data, especially when dealing with visual scenes and image-based data.

Innovation Solution

An image file generating device and method that convert image data into training data by attaching annotation-related information as metadata, allowing users to select and organize their concerns, such as event or timing information, to create effective training data for inference models.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If inference models are generated using deep learning with training data, then various problems can be handled flexibly, but generating training data that aligns with user's intention becomes complicated

Engineering Contradiction:
Improveflexibility in handling various problemsVSAvoidcomplexity of generating training data
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent segments the training data generation process into distinct components: image data acquisition, annotation information attachment, and metadata organization. By dividing the complex task of creating user-aligned training data into manageable segments with standardized formats, the system maintains flexibility in handling various problems while reducing the overall complexity of the generation process

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent changes the parameters of training data generation by introducing standardized annotation formats and metadata structures. This allows the system to maintain adaptability across different problem domains while simplifying the generation process through consistent parameter organization and predefined data schemas

Inventive Principle:
Principle #35Parameter changes

2Ease of manufacture

If standardized methods for creating training data are introduced, then the process becomes simpler, but the ability to capture diverse user concerns may be limited

Engineering Contradiction:
Improvesimplicity of training data creationVSAvoidcoverage of user concerns
Core Design Contradiction:
Ease of manufactureVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal training data structure that can serve multiple functions and capture diverse user concerns. The standardized annotation system is designed to be multi-functional, accommodating various types of information (event detection, timing, object identification) within a single framework, thereby simplifying the creation process while maintaining broad adaptability

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent adds another dimension to training data by introducing metadata layers that organize annotation information. This dimensional addition allows the standardized structure to capture complex user concerns without increasing surface-level complexity, as the metadata hierarchy provides organized access to diverse information types

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentUS11281940B2Image file generating device and image file generating method
Publication Date: 2022.03.22 OLYMPUS CORPORATION(JP)
  • US11281940B2 patent drawing
  • US11281940B2 patent drawing
  • US11281940B2 patent drawing

AI summary

An image file generating device, comprising a processor having an image input section and a file generation section, the image input section inputting image data that has been converted based on images, wherein the file generating section attaches annotation related information to the image data as metadata for a relationship between the image data and user's concern, and makes into a file as training data, wherein the annotation related information includes information that has been selected from within at least one of (1) information as to whether what is being determined is an event or not, and (2) information as to whether what is being determined is timing or not.