Object Detection Result Selection for Faster Image Annotation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI training methods require extensive manual annotation of large datasets, and pre-trained models often present erroneous detection results, leading to inefficiencies and inaccuracies in object detection.
Innovation Solution
An information processing apparatus and method that acquires and selects the most accurate object detection results from multiple pre-trained models, using threshold values and user confirmation to filter out erroneous outputs, and stores the selected results for annotation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all detection results of a pre-trained model are presented to the user, then the user can review all possible annotations, but erroneous detection results are also presented along with correct ones
Solution Approach 1:
A selection unit is introduced as an intermediary component between the pre-trained model and the user interface. This selection unit automatically filters detection results by comparing them against reference data, allowing only high-confidence results to be presented to the user while blocking erroneous detections. This resolves the contradiction by maintaining reliability through automated filtering without losing useful information.
Solution Approach 2:
The system performs self-service by automatically evaluating and selecting its own detection results using the selection unit. The model's outputs are autonomously assessed against reference data, and the system self-regulates which results to present, eliminating the need for manual review of all detections while maintaining high accuracy.
2Manufacturing precision
If manual annotation work is performed for all images, then high quality supervisor data can be generated, but a large amount of time is required
Solution Approach 1:
Pre-trained models perform preliminary annotation work by automatically detecting objects in images before manual review. The selection unit then filters these preliminary results, presenting only high-confidence detections to annotators. This preliminary automated processing significantly reduces the time required for manual annotation while maintaining data quality through selective human review.
Solution Approach 2:
The system creates reference data by copying and storing ground truth annotations from previously annotated images. This reference data is then used by the selection unit to automatically evaluate and filter new detection results, enabling rapid processing of new images without requiring complete manual re-annotation while preserving annotation quality.
3Measurement precision
If multiple pre-trained models are used to improve detection accuracy, then more accurate results can be obtained, but the complexity of the system increases
Solution Approach 1:
Multiple pre-trained models are merged into a unified processing pipeline where their detection results are collectively evaluated by a single selection unit. The selection unit applies consistent filtering criteria to results from all models, combining their strengths while maintaining systematic control. This merging approach improves detection accuracy through model diversity without proportionally increasing system complexity.
Solution Approach 2:
The selection unit serves as a universal component that handles detection results from multiple different pre-trained models using the same filtering mechanism. This multi-functional selection unit can evaluate any model's output against the reference data, allowing the system to leverage multiple models for improved accuracy while avoiding the need for separate processing logic for each model.
Data Source
AI summary
An information processing apparatus comprises an acquisition unit configured to, for each of pre-trained models, acquire results of object detection from an image performed by the pre-trained model, a selection unit configured to, for each of the pre-trained models, select a result to be outputted out of the results of object detection from the image performed by the pre-trained model, and an output unit configured to output the result selected by the selection unit for each pre-trained model.