Container Image Encoding for Compression-Resilient Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating image data for training machine learning models, such as deep neural networks, face challenges with the loss of non-visual information during compression and the need for consistent naming conventions in container files, leading to inefficiencies and increased complexity in data handling and processing.
Innovation Solution
A method that integrates binary and/or text information into a container image using one- or two-dimensional codes, allowing for secure and reliable data exchange by embedding additional data within the image, ensuring compatibility with lossy compression and reducing the need for multiple files.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If additional data are embedded in image data using steganography, then data integrity and reliability are improved, but device complexity and processing difficulty increase
Solution Approach 1:
The patent merges image data and additional data (labels, metadata) into a single container image file. The additional data is encoded and embedded within the image data structure, eliminating the need for separate label files and reducing the complexity of data management while ensuring data integrity through unified storage.
Solution Approach 2:
The container image serves multiple functions: it stores visual image information, embeds additional data (labels, metadata), and maintains compatibility with standard image processing tools. This multi-functional approach reduces the need for specialized processing tools and simplifies the overall system architecture.
2Ease of operation
If container files are used to store image data and labels, then data organization is improved, but consistency in naming conventions and file handling complexity increase
Solution Approach 1:
The patent combines image data and associated labels into a single container image file, eliminating the need for separate label files and reducing file handling complexity. The container image self-contained structure removes the burden of maintaining consistent naming conventions across multiple files.
Solution Approach 2:
The container image is designed to be self-contained, carrying all necessary information (image data, labels, metadata) within its own structure. This self-service capability eliminates the need for external reference files and simplifies data processing workflows.
3Ease of manufacture
If additional data are encoded in additional lines or columns of the image, then data storage is simplified, but data loss occurs during compression
Solution Approach 1:
The patent embeds additional data within the image data structure itself, nesting labels and metadata within the container image. This nested structure allows the additional data to be stored as part of the image data, protecting it from loss during compression while maintaining storage simplicity.
Solution Approach 2:
The patent converts the potential harm of data loss during compression into a benefit by embedding additional data in a way that makes it resilient to compression. The embedded data structure is designed to survive compression processes, transforming the compression challenge into an opportunity for more compact storage.
4Adaptability or versatility
If separate label files are used, then data processing flexibility is improved, but data exchange reliability and processing efficiency decrease
Solution Approach 1:
The patent merges image data and labels into a single container image file, improving processing efficiency by eliminating the need to handle multiple separate files. The unified structure maintains data exchange reliability while streamlining processing workflows.
Solution Approach 2:
The patent performs preliminary organization of data by pre-embedding labels and metadata within the container image structure. This preliminary action eliminates the need for subsequent file association and processing steps, improving overall processing efficiency while maintaining flexibility through the structured data organization.
Data Source
AI summary
The invention relates to a method for generating image data that are enhanced with at least one piece of binary and/or text information. The invention further relates to a data structure, a computer program, a device, and a memory medium.