Unified R-CNN Network for Multi-Class Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object detectors based on R-CNNs require multiple deep learning networks for different data sets, leading to increased execution time in real-world tests, as they need separate networks for each class of objects, making them inefficient for detecting multiple objects like pedestrians, riders, cars, traffic signs, and animals simultaneously.
Innovation Solution
A method is introduced where a single R-CNN network learns parameters using multiple data sets by employing a learning device that instructs convolutional layers to generate feature maps, ROI regression information, and matching information, allowing the network to output ROI proposals and feature vectors for multiple object classes, and backpropagates losses to learn parameters efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple separate R-CNN networks are used for different object classes, then each network can be specialized for its specific data set, but the execution time increases in proportion to the number of networks
Solution Approach 1:
The patent merges multiple separate R-CNN networks into a single unified R-CNN network that can process multiple object classes simultaneously. The network architecture integrates multiple convolutional layers, ROI pooling layers, and fully connected layers that handle different object classes (pedestrians, riders, cars, traffic signs, traffic lights, animals) in parallel within one network, eliminating the need to sequentially execute multiple separate networks.
Solution Approach 2:
The unified R-CNN network is designed with multi-functionality to detect various types of objects including pedestrians, riders, cars, traffic signs, traffic lights, and animals within a single network. The network uses shared convolutional feature extraction layers followed by class-specific ROI pooling and classification layers, allowing one network to perform the functions previously requiring multiple specialized networks.
2Reliability
If multiple separate R-CNN networks are trained for different object classes, then each network learns from its specific data set, but the device complexity increases with the number of networks
Solution Approach 1:
The patent combines multiple separate detection networks into a single integrated network architecture. Instead of maintaining separate R-CNN networks for pedestrians, riders, cars, traffic signs, traffic lights, and animals, the invention creates one unified network that processes all object classes through shared feature extraction layers, reducing the overall number of network components and simplifying the system architecture.
Solution Approach 2:
The unified network employs a universal feature extraction backbone that serves all object classes, with subsequent branching into class-specific detection pathways. This multi-functional design allows a single network structure to handle diverse object types, reducing device complexity compared to maintaining multiple separate specialized networks.
3Reliability
If separate data sets are prepared for each object class, then each data set can be optimized for its specific class, but the preparation complexity and time increase
Solution Approach 1:
The patent merges the requirement for multiple separate data sets into a single unified data set structure. Instead of preparing and managing separate data sets for pedestrians, riders, cars, traffic signs, traffic lights, and animals, the invention consolidates all training data into one data set that feeds the unified network, simplifying data collection, annotation, and management processes while maintaining class-specific detection capabilities.
Data Source
Figure 1(A)~1(C)
Figure 2
Figure 3
AI summary
A method for learning parameters of an object detector based on R-CNN is provided. The method includes steps of: a learning device (a) if training image is acquired, instructing (i) convolutional layers to generate feature maps by applying convolution operations to the training image, (ii) an RPN to output ROI regression information and matching information (iii) a proposal layer to output ROI candidates as ROI proposals by referring to the ROI regression information and the matching information, and (iv) a proposal-selecting layer to output the ROI proposals by referring to the training image; (b) instructing pooling layers to generate feature vectors by pooling regions in the feature map, and instructing FC layers to generate object regression information and object class information; and (c) instructing first loss layers to calculate and backpropagate object class loss and object regression loss, to thereby learn parameters of the FC layers and the convolutional layers.