Dynamic Neural Network Scheduling for Visual Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer vision methods face challenges in achieving a balance between tracking accuracy and computational efficiency, particularly in resource-constrained devices, due to the use of fixed neural network architectures that either consume excessive resources or compromise on tracking precision.
Innovation Solution
Dynamic neural network scheduling that selects the most suitable neural network from a pool based on immediate tracking requirements and available resources, adapting to environmental changes and resource availability to maintain accuracy while optimizing computational efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a fixed neural network architecture is used, then the system structure is simple, but computational resource consumption is excessive or tracking precision is compromised
Solution Approach 1:
The patent applies dynamics by transitioning from a fixed neural network architecture to a dynamic scheduling system that selects different neural networks based on real-time tracking requirements and resource availability. The system dynamically switches between neural networks with different computational complexities to optimize the balance between tracking precision and resource consumption.
Solution Approach 2:
The patent changes the parameter of neural network architecture selection based on varying conditions. The scheduler monitors tracking performance metrics and resource availability, then adjusts which neural network is active accordingly. This parameter change allows the system to adapt computational resource consumption to actual needs.
2Device complexity
If a fixed neural network architecture is used, then the system structure is simple, but tracking precision is compromised
Solution Approach 1:
The system dynamically selects neural networks based on real-time tracking requirements. When high tracking precision is needed, the scheduler activates more accurate neural networks; when precision requirements are lower, it switches to less computationally intensive networks. This dynamic adaptation maintains tracking precision without requiring a permanently complex system architecture.
Solution Approach 2:
The patent changes the neural network architecture parameter based on tracking precision requirements. The scheduler monitors performance metrics and adjusts which neural network is active, allowing the system to optimize tracking precision when needed while maintaining simplicity during less critical periods.
3Measurement precision
If a high-accuracy neural network is used, then tracking precision is improved, but computational resource consumption increases
Solution Approach 1:
The patent implements dynamic scheduling that matches neural network complexity to actual tracking needs. The system monitors tracking performance and resource availability in real-time, switching between high-accuracy and resource-efficient neural networks as conditions change. This ensures high tracking precision is achieved only when and where needed, avoiding unnecessary computational resource consumption.
Solution Approach 2:
The system changes the neural network architecture parameter dynamically based on the trade-off between tracking precision and resource consumption. The scheduler adjusts which neural network is active according to current requirements, allowing the system to optimize the balance between achieving high tracking precision and conserving computational resources.
4Use of energy by moving object
If a low-accuracy neural network is used, then computational resource consumption is reduced, but tracking precision is compromised
Solution Approach 1:
The dynamic scheduling system allows the system to switch between low-resource and high-precision neural networks based on real-time conditions. When computational resources are constrained, the system uses lighter neural networks; when precision is critical, it activates more accurate networks. This dynamic adaptation ensures tracking precision is maintained when needed without permanently consuming excessive computational resources.
Solution Approach 2:
The patent changes the neural network architecture parameter based on the balance between resource consumption and tracking precision requirements. The scheduler monitors system state and adjusts which neural network is active, allowing the system to optimize resource usage while maintaining acceptable tracking precision through intelligent parameter selection.
Data Source
AI summary
An example apparatus includes interface circuitry, machine-readable instructions, and at least one processor circuit to be programmed by the machine-readable instructions to identify at least one region of interest associated with a computer vision task, measure accuracy of the computer vision task in a covariance space, the accuracy of the computer vision task associated with computer resource consumption, and select at least one neural network from a group of neural networks based on the measured accuracy of the computer vision task.