Object Detection Model Training via Reliability-Based Area Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current robot systems face challenges in accurately detecting areas containing target objects from input images, leading to inefficiencies in object detection and identification processes.
Innovation Solution
An information processing apparatus comprising a first detection unit for detecting areas within input images, an identification unit for calculating feature vectors and determining identification reliability, and a learning unit that selects and learns from detection areas based on identification reliability to enhance object detection models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If detection processing is performed on all input images to identify objects, then object identification capability is improved, but processing time and computational load increase
Solution Approach 1:
The system performs detection processing selectively rather than on all images. The learning unit identifies detection areas with high identification reliability and uses only those for model learning, performing partial processing instead of exhaustive processing on all input images, thereby reducing overall processing time while maintaining identification accuracy
Solution Approach 2:
The system pre-processes images to identify detection areas with high identification reliability before performing full object identification. By preliminarily selecting reliable detection areas and using them for model learning, the system prepares optimized detection models that can quickly and accurately identify objects in future images without requiring exhaustive processing of all possible detection areas
2Reliability
If detection processing is performed on all input images to identify objects, then object identification capability is improved, but computational resources consumed increase
Solution Approach 1:
The system performs detection processing only on selected detection areas rather than all input images. The learning unit identifies and processes only those detection areas with high identification reliability, significantly reducing the number of computations required while maintaining or improving object identification accuracy through targeted learning
Solution Approach 2:
The system extracts and isolates only the high-reliability detection areas from the full set of input images for model learning. By separating and processing only the relevant detection areas identified by the learning unit, the system eliminates unnecessary computational operations on low-reliability areas, reducing overall computational load and energy consumption
3Productivity
If a simple detection model is used, then processing speed is improved, but detection accuracy deteriorates
Solution Approach 1:
The system pre-trains detection models using carefully selected high-reliability detection areas identified by the learning unit. This preliminary training on quality data creates optimized models that achieve high detection accuracy. Once trained, these models can perform rapid detection without requiring complex real-time processing, thus achieving both speed and accuracy
Solution Approach 2:
The system changes the quality parameters of training data by selecting detection areas with high identification reliability. By adjusting the parameter of data quality (selecting only high-reliability detection areas for model training), the system creates more effective detection models that achieve higher accuracy with simpler architectures and faster processing speeds
Data Source
AI summary
An area including an object is detected from an input image with high accuracy. An information processing apparatus of the present disclosure includes: a first detection unit that performs detection processing of detecting an area including an object with respect to an input image; an identification unit that calculates a feature vector on the basis of an image of an area detected in the detection processing, identifies the object on the basis of the feature vector, and acquires identification reliability that is reliability of an identification result of the object; and a learning unit that selects a detection area for learning from a plurality of detection areas corresponding to a plurality of the feature vector on the basis of a plurality of the identification reliability, and learns a model that detects an area including the object on the basis of an image of the detection area selected.