Descriptor Image Generation Using Segmented Machine Learning Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for training machine learning models to generate descriptor images for objects are inefficient and lead to inconsistencies when expanding the class of objects, requiring complete retraining and resulting in loss of previous training results.
Innovation Solution
Training multiple machine learning models for different object classes, allowing for the expansion of manageable objects without retraining previous classes, using a method that evaluates distances between reference descriptors and generated descriptors to assign accurate descriptor images, and utilizing shared backbone networks to reduce training effort.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a machine learning model is trained for a first class of objects and then retrained for a second class of objects, then the model can provide descriptors for the second class, but the original training for the first class is lost and training effort increases
Solution Approach 1:
The patent divides the single machine learning model into multiple specialized models, with each model trained for a specific object class. This segmentation allows each model to maintain specialized knowledge for its designated class while avoiding the need to retrain the entire model when encountering new object classes, thus reducing training effort and preserving previous training results.
Solution Approach 2:
The patent creates a system where multiple machine learning models work together to handle different object classes. Each model is specialized for a specific class but all models collectively provide universal coverage across multiple object classes. This multi-functionality approach allows the system to handle diverse object classes without requiring complete retraining, as each specialized model can independently process its designated class.
2Adaptability or versatility
If a machine learning model is retrained to include new object classes, then the class of objects can be expanded, but inconsistencies in descriptor ascertainment occur
Solution Approach 1:
By segmenting the machine learning models into separate specialized models for different object classes, the patent ensures that each model maintains consistent descriptor generation for its designated class. This segmentation prevents the inconsistencies that would arise from retraining a single model to handle multiple classes, as each specialized model preserves its original training consistency while the system as a whole expands to handle multiple object classes through model selection.
3Adaptability or versatility
If complete retraining is performed to expand object classes, then all object classes can be handled, but training effort and computational resources increase significantly
Solution Approach 1:
The patent segments the training process into separate specialized models for different object classes. When expanding to new object classes, only the necessary specialized models need to be trained or activated, rather than performing complete retraining of all models. This segmentation significantly reduces computational resources and training effort required to expand the system's capability to handle new object classes.
Solution Approach 2:
Instead of performing complete retraining of the entire model system when expanding object classes, the patent applies partial action by training only the specific specialized models needed for the new object classes. This partial training approach reduces computational resources and training time while still achieving the goal of expanding object class handling capability.
Data Source
AI summary
A method for ascertaining a descriptor image for an image of an object. The method includes training, for each of a plurality of object classes, a respective machine learning model to map images of objects of the object class to descriptor images and storing reference descriptors output by the machine learning model for one or more objects of the object class; receiving an image of an object; generating, for each object class, a respective descriptor image for the object by mapping the received image to a descriptor image using the machine learning model trained for the object class; evaluating, for each object class, the distance between the reference descriptors stored for the object class and the descriptors of the descriptor image generated for the object class; and assigning the descriptor image to the object as the descriptor image of the object generated for an object class based on the distance.