Extended-Class Object Detection Using Cross-Annotated Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for training object detectors to detect extended object classes require significant time and cost due to the need for extensive annotation of new images and classes, as existing training data sets are insufficient for accurately identifying new object classes.
Innovation Solution
A method and device for generating an object detector that uses existing trained detectors to annotate and generate new training data sets, allowing for the detection of both original and new object classes by inputting images into each other's detectors to produce integrated detection results, thereby reducing the need for extensive re-annotation and new image collection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If new images are collected and manually annotated for new object classes, then the object detector can detect extended object classes, but annotation time and cost increase significantly
Solution Approach 1:
The system performs preliminary detection using existing trained detectors to identify potential new objects in existing images before annotation. This preliminary action pre-screens images and identifies regions of interest, so that when new object classes need to be detected, the annotation process only needs to focus on pre-identified regions rather than manually checking entire images, significantly reducing annotation time
Solution Approach 2:
The system introduces existing trained object detectors as intermediaries to assist in the annotation process. These detectors act as mediators that automatically identify and locate new objects in images, providing preliminary annotation results that human annotators can verify and refine, thereby reducing the manual annotation workload and time required
2Adaptability or versatility
If new images are collected and manually annotated for new object classes, then the object detector can detect extended object classes, but annotation cost increases significantly
Solution Approach 1:
The system enables self-service annotation by allowing existing trained detectors to automatically annotate new objects in images. The detectors serve themselves by identifying and marking new object classes without requiring extensive human intervention, thus reducing annotation costs while maintaining detection capability for extended object classes
Solution Approach 2:
The system uses copying by replicating the detection capabilities of existing trained detectors to identify new object classes. The same detection algorithms and models are applied to detect both original and new object classes, allowing the system to leverage existing detection expertise without incurring additional annotation costs for each new class
3Loss of time
If existing training data set is used without additional annotation, then annotation time and cost are reduced, but the object detector cannot accurately detect new object classes
Solution Approach 1:
The system performs preliminary detection using existing trained detectors to identify potential new objects and their locations in existing images. This preliminary action creates a foundation of pre-identified regions that can be directly used for training new object class detectors, eliminating the need for time-consuming manual annotation while ensuring detection accuracy through pre-screened, high-quality candidate regions
Solution Approach 2:
The system applies partial annotation by only annotating regions identified by existing detectors rather than annotating entire images or all possible regions. This partial action focuses annotation efforts only on relevant areas, reducing overall annotation time and cost while maintaining sufficient data quality and quantity for accurate detection of new object classes
Data Source
AI summary
A method for generating an object detector based on deep learning capable of detecting an extended object class is provided. The method is related to generating the object detector based on the deep learning capable of detecting the extended object class to thereby allow both an object class having been trained and additional object class to be detected. According to the method, it is possible to generate the training data set necessary for training an object detector capable of detecting the extended object class at a low cost in a short time and further it is possible to generate the object detector capable of detecting the extended object class at a low cost in a short time.