Modified Image Representations for Training Data Evaluation

Solution Overview

Problem

The existing methods for training computer vision applications in manufacturing environments are time-consuming and bandwidth-intensive due to the need to transmit and process large high-resolution images, which can be inefficient and tedious, especially when evaluating thousands of images for suitability in training processes.

Innovation Solution

A system that generates and transmits modified image representations with reduced file sizes, such as down-sampled or compressed versions, allowing users to evaluate and select images for training purposes efficiently, while the server generates high-resolution images only when deemed necessary, thereby reducing bandwidth and processing time.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If large high-resolution images are transmitted for training computer vision applications, then training accuracy is improved, but bandwidth consumption and processing time increase significantly

Engineering Contradiction:

Improvetraining accuracyVSAvoidprocessing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent segments the image processing workflow into two distinct phases: evaluation phase using down-sampled representations and training phase using full-resolution images. This segmentation allows the system to use low-resolution images for initial evaluation and selection, then only transmit full-resolution images for actual training, thereby reducing overall processing time while maintaining training accuracy.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent creates down-sampled copy representations of original high-resolution images. These copies serve as proxies for evaluation purposes, allowing users to assess image quality and suitability without handling the full-resolution originals. This copying approach significantly reduces bandwidth consumption during the evaluation and selection phases while preserving the ability to use originals when needed for training.

Inventive Principle:

Principle #26Copying

2Measurement precision

If large high-resolution images are transmitted for training computer vision applications, then training accuracy is improved, but bandwidth consumption increases significantly

Engineering Contradiction:

Improvetraining accuracyVSAvoidbandwidth consumption

Core Design Contradiction:

Measurement precisionVSLoss of energy

Solution Approach 1:

The patent segments the image processing workflow into two distinct phases: evaluation phase using down-sampled representations and training phase using full-resolution images. This segmentation allows the system to use low-resolution images for initial evaluation and selection, then only transmit full-resolution images for actual training, thereby reducing overall processing time while maintaining training accuracy.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent creates down-sampled copy representations of original high-resolution images. These copies serve as proxies for evaluation purposes, allowing users to assess image quality and suitability without handling the full-resolution originals. This copying approach significantly reduces bandwidth consumption during the evaluation and selection phases while preserving the ability to use originals when needed for training.

Inventive Principle:

Principle #26Copying

3Reliability

If thousands of images are evaluated for suitability in training processes, then comprehensive training data selection is improved, but the evaluation process becomes increasingly tedious and time-consuming

Engineering Contradiction:

Improvetraining data selection qualityVSAvoidevaluation ease

Core Design Contradiction:

ReliabilityVSEase of operation

Solution Approach 1:

The patent creates down-sampled copy representations of original high-resolution images. These copies serve as proxies for evaluation purposes, allowing users to assess image quality and suitability without handling the full-resolution originals. This copying approach significantly reduces bandwidth consumption during the evaluation and selection phases while preserving the ability to use originals when needed for training.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent changes the resolution parameter of images during different phases of the process. Down-sampled representations are used for initial evaluation to reduce visual complexity and processing demands, while full-resolution images are used for training when needed. This parameter transformation makes the evaluation of thousands of images more manageable and less tedious while maintaining the ability to assess training data quality comprehensively.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20240289918A1Representative modified images

Publication Date: 2024.08.29 FORD GLOBAL TECH LLC

AI summary

Generating modified image representations may utilize a system, which includes a computer having a processor coupled to a memory, the memory including instructions executable by the processor, to obtain user input of parameters to direct probabilities of modification types for a base image stored in a database, and to generate, responsive to the user input, a representation of the base image that includes modifications according to the parameters. The processor coupled to the memory may additionally direct the computer to transmit the representation of the base image to a second computer for display.