Unrecognized Object Memory Mechanism for Image Model Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image object detection technologies struggle to recognize and categorize unrecognized objects without requiring extensive manual labeling and costly training data, and semantic segmentation algorithms fail to identify object categories.
Innovation Solution
An image analyzation method with an unrecognized object memory mechanism that groups images with similar features into specified directories and assigns new category labels when a threshold is reached, using a feature extraction and classification process to enable the object recognition model to learn and recognize new objects automatically.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a large-scale object recognition model is trained to recognize new object categories, then the model's ability to identify unrecognized objects is improved, but the cost and workload of collecting and labeling training data increases significantly
Solution Approach 1:
The system enables the object recognition model to perform self-learning by automatically collecting unlabeled images, generating pseudo-labels through its own predictions, and retraining itself without requiring manual data annotation. This self-service mechanism allows the model to adapt to new object categories while avoiding the high costs of traditional supervised learning data preparation
Solution Approach 2:
The system implements a feedback loop where the object recognition model's predictions are continuously evaluated, and incorrect or low-confidence predictions trigger automatic collection of additional training data. The model uses this feedback to iteratively improve its recognition capability for unrecognized objects, creating a closed-loop learning system that adapts over time
2Ease of operation
If semantic segmentation algorithms are used to separate objects from background, then object separation is achieved, but the ability to identify object categories and create associations among objects is lost
Solution Approach 1:
The system merges the strengths of both object recognition and semantic segmentation by integrating their outputs. It uses segmentation masks to accurately separate objects from backgrounds while simultaneously applying recognition models to identify object categories, thereby preserving both separation quality and category information in a unified framework
3Adaptability or versatility
If the object recognition model is retrained with new training data, then the model recognizes new object categories, but the training time and computational resources increase
Solution Approach 1:
Instead of completely retraining the model with all available data, the system performs partial retraining by selectively updating only the specific object categories that need improvement. This partial action approach reduces training time and computational resources while still achieving the goal of recognizing new object categories
Data Source
AI summary
An image analyzation method with an unrecognized object memory mechanism are provided. Initially, whether a target object belongs to a known category using an object recognition model is determined; if not, then the object memory mechanism is executed. The mechanism first obtains image feature information through a feature extraction algorithm, and generates an attention representative value by a feature classifier. The target objects are grouped according to the numerical range it belongs to and stored in corresponding specified directories. When the number of target images in a directory reaches a quantity threshold, a new category label is assigned to the target images and used as retraining data for the object recognition model. The image analyzation method classifies unrecognized objects and enables the object recognition model to recognize unrecognized objects through learning, without the need for manual labeling of unrecognized objects, thereby saving resources and cost.