Neural Network Inference Parameter Switching for Image Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image pickup apparatuses, such as digital cameras, do not efficiently utilize dedicated circuits for neural network processing, particularly when switching between different inference parameters for shooting control and image processing control.
Innovation Solution
An information processing apparatus with a processor that sets and switches between first and second inference parameters in response to focus target settlement, allowing efficient neural network switching for improved processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a dedicated circuit with multiply-accumulate operation is used for neural network processing, then processing speed and efficiency are improved, but the ability to switch between different inference parameters for different processing tasks becomes insufficient
Solution Approach 1:
The patent implements dynamic switching between first inference parameters (for shooting control) and second inference parameters (for image processing control) based on processing stage requirements. The system transitions from static to dynamic parameter configuration, allowing the same dedicated circuit to adaptively serve different neural network processing tasks at different times.
Solution Approach 2:
The patent changes the inference parameters of the neural network based on the processing stage. First inference parameters are used during shooting control phase, and second inference parameters are used during image processing control phase. This parameter switching enables the dedicated circuit to handle different types of neural network processing efficiently.
2Device complexity
If the same dedicated circuit is used for both shooting control and image processing control, then device complexity is reduced, but processing delays occur due to sequential operation
Solution Approach 1:
The patent prepares and switches inference parameters in advance based on predicted processing needs. By determining which inference parameters will be needed next and switching to them proactively, the system reduces waiting time and avoids processing delays while maintaining a single dedicated circuit.
Solution Approach 2:
The patent implements periodic switching between first and second inference parameters according to the processing stage. The system alternates between shooting control mode and image processing control mode in a structured periodic manner, allowing the dedicated circuit to efficiently handle different tasks sequentially with minimal idle time.
3Productivity
If multiple dedicated circuits are used for different inference parameters, then processing efficiency is improved, but device complexity and cost increase
Solution Approach 1:
The patent makes a single dedicated circuit universal by enabling it to perform both shooting control processing and image processing control processing through inference parameter switching. This multi-functionality approach achieves parallel processing capabilities without requiring separate dedicated circuits for each task, thus avoiding increased device complexity.
Data Source
AI summary
An information processing apparatus includes a processor that performs arithmetic processing of a neural network, and a controller that is capable of setting, in the processor, a first inference parameter to be applied to processing for shooting control for shooting an image and a second inference parameter to be applied to processing for processing control of the image. The controller switches the first inference parameter having been set in the processor to the second inference parameter in response to settlement of a focus target.