Image Labeling System Combining Specific and Generic Modules
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image labeling systems face challenges in reusing specialized modules trained on specific labels for labeling images with different labels, leading to difficulties in achieving efficient labeling performance.
Innovation Solution
A method that combines specific and generic descriptive data from convolutional neural networks to create an efficient image labeling system, where the first module is trained on a first learning corpus and the second module is trained on a second learning corpus with generic labels, and both modules are used to provide descriptive data to a downstream module for improved labeling performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a first labeling module is trained on a first training corpus with specific labels, then the module can accurately label images with those specific labels, but the module cannot efficiently label images with different labels
Solution Approach 1:
The patent segments the labeling module into two independent components: a first labeling module trained on specific labels and a second labeling module trained on generic labels. Each module processes images independently and their outputs are combined, allowing the system to maintain specialization while gaining versatility for different labeling tasks.
Solution Approach 2:
The patent creates a universal labeling system where the second labeling module, trained on generic labels, can handle multiple different labeling tasks. This module serves as a versatile component that adapts to various label sets while the first module provides specialized accuracy for specific labels.
2Measurement precision
If only specific descriptive data from the first labeling module is used, then labeling accuracy for specific labels is maintained, but the system cannot efficiently adapt to new labels
Solution Approach 1:
The patent merges the outputs of two labeling modules by combining their descriptive data through a combination module. This integration allows the system to leverage both the specific label accuracy from the first module and the adaptability to new labels from the second module, achieving both precision and efficiency simultaneously.
3Adaptability or versatility
If only generic descriptive data from the second labeling module is used, then the system can handle different labels efficiently, but labeling accuracy decreases
Solution Approach 1:
The patent introduces a combination module as an intermediary that processes and integrates descriptive data from both labeling modules. This intermediary component ensures that the generic descriptive data from the second module is properly combined with specific descriptive data from the first module, maintaining both flexibility and accuracy in the final labeling output.
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
This method comprises: the obtaining (102) of a first module for labelling images by automatic training on the basis of a first training corpus; the obtaining (104) of a second training corpus on the basis of the first training corpus, by replacing, in the first training corpus, each of at least one part of first labels by a replacement label, at least two first labels being replaced with one and the same replacement label; the obtaining (106) of a second module for labelling images by automatic training on the basis of the second training corpus; the obtaining (108) of the system for labelling images comprising: a first upstream module obtained on the basis of at least one part of the first module for labelling images, a second upstream module obtained on the basis of at least one part of the second module for labelling images and a downstream module designed to provide a labelling of an image on the basis of first descriptive data provided by the first upstream module and of second descriptive data provided by the second upstream module.