Class-Focused Segmentation Training for Compact Edge Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image segmentation models are not suitable for on-device applications due to their large parameter count and resource requirements, and they struggle with the diversity of agricultural domains, requiring extensive training data and platform constraints.
Innovation Solution
A two-stage model training approach is employed, first creating a master ML segmentation model and then a class-focused segmentation model using a reduced search space and hyperparameters, followed by fine-tuning on task-specific datasets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing image segmentation models are used, then segmentation accuracy can be achieved, but the model size and resource requirements become too large for on-device deployment
Solution Approach 1:
The training process is divided into two distinct stages: a first stage that trains on a master dataset with multiple classes to learn general features, and a second stage that fine-tunes on task-specific datasets with fewer classes to specialize in specific segmentation tasks. This segmentation of the training process enables the model to achieve high accuracy while maintaining compact size for on-device deployment.
Solution Approach 2:
The system dynamically adjusts model parameters and architecture based on the specific task requirements. By receiving task-specific datasets and focusing training on relevant classes, the model adapts its parameters to achieve optimal performance for each specific segmentation task while maintaining a compact form factor suitable for edge devices.
2Measurement precision
If models are trained for each specific agricultural domain, then domain accuracy improves, but the process becomes non-scalable and time-consuming
Solution Approach 1:
A master model is pre-trained on a comprehensive master dataset that encompasses multiple agricultural domains and classes. This preliminary training establishes a strong foundation of general agricultural knowledge that can be quickly adapted to specific domains through the second stage of task-specific fine-tuning, significantly reducing the time required for domain-specific model development.
Solution Approach 2:
The master model is designed to be universal across multiple agricultural domains and crop types. By training on a diverse master dataset that includes various classes from different agricultural contexts, the model learns transferable features that can be applied across multiple domains, making the system scalable and reducing the need for separate models for each specific agricultural application.
3Loss of time
If AutoML tools are used with limited training data, then the process is faster, but the models become overfitted and poor in quality
Solution Approach 1:
The master model is pre-trained on a large, diverse master dataset before being fine-tuned on smaller task-specific datasets. This preliminary action ensures that the model learns robust general features from abundant data, preventing overfitting when subsequently trained on limited task-specific data. The two-stage approach allows rapid adaptation to specific tasks while maintaining model quality through the foundation established in the first stage.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Existing segmentation models have the disadvantage that they cannot be deployed on edge devices as they are not compact and require large space. Embodiments disclosed herein provide a method and system for generating a class focused segmentation model using a two stage model training approach, which enables generation of a compact, task specific model. Using a first stage of training, a Machine Learning (ML) segmentation model that is trained on a master training dataset is generated. Further, in a second stage of training, a fine-tuned task specific segmentation model is generated, which can be used for task specific class segmentation. The fine-tuned task specific segmentation model being task specific and in turn compact, can be used edge device deployment and such applications.