Multi-Scale Region Proposal Network for Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object detection methods in computing devices face challenges in simultaneously performing multi-class recognition and accurate localization, as they often require separate processing of region proposals and object classification, leading to inefficiencies and reduced accuracy.
Innovation Solution
A computing device is designed with a multi-layer convolution network and a multi-scale region proposal network that generates region proposals with varying receptive field sizes, combined with a multi-scale object classifier to classify objects at multiple scales, reducing computational costs and preventing duplication of pooled features for small objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If classical approaches separate object localization and classification tasks, then each task can be processed independently, but a large number of candidate object locations must be processed before classifying, leading to reduced efficiency and accuracy
Solution Approach 1:
The patent combines object localization and classification tasks into a unified deep learning framework where region proposal networks and object classifiers work together in an integrated architecture, eliminating the need for separate processing stages and reducing overall system complexity while improving efficiency
2Use of energy by moving object
If a single network is used for object detection, then computational cost is reduced by sharing front-end convolutional layers, but accuracy is lost for objects of varying scales
Solution Approach 1:
The patent segments the object detection network into multiple pathways with different receptive field sizes, allowing each pathway to specialize in detecting objects at specific scales while sharing common front-end convolutional layers, thus maintaining computational efficiency while improving multi-scale detection accuracy
Solution Approach 2:
The patent introduces multi-scale processing by adding the dimension of varying receptive field sizes to the network architecture, enabling simultaneous processing of objects at different scales through parallel convolutional pathways with different field-of-view characteristics
3Device complexity
If region proposals are generated with fixed receptive field size, then the network structure is simpler, but accuracy is reduced for objects of different scales
Solution Approach 1:
The patent implements dynamic receptive field selection by generating region proposals with multiple different receptive field sizes based on the detected object scale, allowing the network to adapt its processing parameters dynamically rather than using fixed parameters throughout
Data Source
AI summary
Technologies for multi-scale object detection include a computing device including a multi-layer convolution network and a multi-scale region proposal network (RPN). The multi-layer convolution network generates a convolution map based on an input image. The multi-scale RPN includes multiple RPN layers, each with a different receptive field size. Each RPN layer generates region proposals based on the convolution map. The computing device may include a multi-scale object classifier that includes multiple region of interest (ROI) pooling layers and multiple associated fully connected (FC) layers. Each ROI pooling layer has a different output size, and each FC layer may be trained for an object scale based on the output size of the associated ROI pooling layer. Each ROI pooling layer may generate pooled ROIs based on the region proposals and each FC layer may generate object classification vectors based on the pooled ROIs. Other embodiments are described and claimed.