Modular Trained Model Generation for Robot Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating trained models for robot tasks, such as object recognition, face challenges in increasing inference accuracy and efficiency, particularly in adapting to different environments and recognizing specific targets within large categories.
Innovation Solution
The approach involves generating a trained model by connecting and training multiple preliminary models, each with a backbone and a head, using different sets of information to enhance recognition accuracy and adaptability, allowing for the creation of both general-purpose and dedicated models for various recognition tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple base models are trained separately and then combined, then recognition accuracy for various targets is improved, but device complexity and training time increase
Solution Approach 1:
The trained model is segmented into multiple independent base models, each trained on specific training data for particular recognition targets. This allows each base model to specialize in recognizing certain targets while maintaining overall system accuracy. The segmentation enables parallel training of multiple models rather than training one large complex model.
Solution Approach 2:
The system creates a universal trained model structure that can perform multiple recognition functions through different base models. Each base model contributes to the overall recognition capability, making the system multi-functional. The trained model can recognize various targets including industrial parts and other objects through its multiple base models.
2Adaptability or versatility
If multiple base models are trained separately using different training data, then adaptability to different environments is improved, but loss of time in training increases
Solution Approach 1:
Multiple base models are trained in advance using different training data sets before being combined into the final trained model. This preliminary action allows each base model to be pre-adapted to specific environments or target types, reducing the need for retraining when deploying to different recognition tasks.
Solution Approach 2:
Multiple base models trained on different training data are merged into a single trained model that combines their recognition capabilities. This merging process integrates the adaptability learned from various training data sets, allowing the final model to handle diverse recognition tasks efficiently.
3Device complexity
If a single trained model is used for all recognition tasks, then device complexity is reduced, but recognition accuracy for specific targets decreases
Solution Approach 1:
Different base models within the trained model have specialized knowledge for specific recognition targets, providing local quality optimization. Each base model is trained on specific training data relevant to particular targets, ensuring high recognition accuracy for those specific targets while maintaining overall system simplicity through unified architecture.
Data Source
AI summary
A trained model generation method includes generating a trained model that outputs a recognition result of a recognition target included in input information, based on multiple models each including at least one of a first portion or a second portion. In the generating of the trained model, multiple base models each including a portion corresponding to the first portion are acquired, the multiple base models being trained based on at least one set of first information related to the input information. In the generating of the trained model, multiple target models each including a portion corresponding to the second portion are acquired, each of the multiple target models being trained based on at least one set of second information related to the input information with being connected to a respective base model of the multiple base models. In the generating of the trained model, the trained model at least including, among the multiple target models, a target model including the portion corresponding to the second portion is generated.