Iterative Training Device for Reducing Manual Annotation Workload
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The high workload associated with manually generating training data for instance segmentation processing is a significant challenge.
Innovation Solution
A training device that iteratively processes image data through a series of trained models, generating and accumulating training data by inputting image data groups into progressively refined models, reducing the need for manual ground truth data assignment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual generation of training data is performed, then ground truth data can be accurately assigned, but the workload of the worker becomes high
Solution Approach 1:
The system uses the trained model to automatically generate ground truth data from image data groups, allowing the model to serve itself in the data generation process. The collection unit collects output image data groups from the trained model, and the generation unit processes this output into ground truth data, eliminating the need for manual annotation while maintaining accuracy.
Solution Approach 2:
The system implements a feedback loop where the trained model processes image data groups to generate output image data groups, which are then used as ground truth data for retraining. This iterative feedback process continuously improves the model's accuracy while automating the data generation process.
2Quantity of substance
If multiple image data groups are processed iteratively, then training data quantity increases, but the complexity of the training process increases
Solution Approach 1:
The system segments the training process into distinct modules: a collection unit that handles image data groups, a generation unit that processes output into ground truth data, and a training unit that performs model training. This segmentation allows each module to handle specific tasks independently, making the overall complex process more manageable and scalable.
Solution Approach 2:
The system performs preliminary processing by collecting output image data groups from the trained model before using them as ground truth data. This preliminary action of gathering and preparing data in advance simplifies the subsequent training process and allows for more efficient iterative processing.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A work load of a worker when generating training data is reduced. A training device includes a training unit configured to perform a training process by using training data including image data included in a default image data group and ground truth data in a case where the image data included in the default image data group is subjected to segmentation processing, to generate a trained model; a collection unit configured to collect an xth output image data group output by inputting an xth image data group among a plurality of image data groups into an (x-1)th-order trained model; and a generation unit configured to generate xth-order training data by acquiring a processed xth output image data group in which each piece of output image data included in the collected xth output image data group is processed into ground truth data, and adding the processed xth output image data group to (x-1)th-order training data. The training unit performs a training process by using the xth-order training data to generate an xth-order trained model.