CAM-Based Weakly Supervised Object Localization via Feature Map Binarization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing weakly supervised object localization methods face challenges in capturing the entire object region due to their reliance on classification tasks, which often focus on small singular regions, and require additional training and computing resources.
Innovation Solution
A CAM-based weakly supervised object localization method that binarizes the weight vector and feature map of a convolutional neural network to generate a class activation map, allowing for selective channel usage and improved object localization without additional learning or computing power.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing CAM generation method is used (multiplying and adding last layer feature map and classifier weight), then classification performance is improved, but object localization performance deteriorates because the network focuses on small singular regions
Solution Approach 1:
The method segments the feature map into multiple channels and selectively processes each channel through binarization. By dividing the feature representation into discrete channel components and applying binary masks, the network can focus on different spatial regions independently, thereby capturing the entire object rather than just singular points while maintaining classification accuracy.
Solution Approach 2:
The invention changes the parameter representation by binarizing the feature map values (converting continuous values to 0 or 1) and applying binary masks to channel selections. This parameter transformation allows the network to maintain classification performance while improving localization by emphasizing regional patterns over point-specific features.
2Measurement precision
If additional training is performed to improve object localization, then localization performance is improved, but computing resources and time increase
Solution Approach 1:
The method performs preliminary binarization of the feature map and channel selection masks before the final CAM generation step. By pre-processing the feature representations into binary form, the network avoids the need for additional training iterations to learn localization patterns, thereby improving localization performance without increasing computing resources or training time.
Solution Approach 2:
The invention creates a binary copy of the feature map that preserves the essential spatial patterns needed for localization while using minimal computational resources. This binary representation serves as a lightweight alternative to full-precision feature maps, enabling improved localization without the computational cost of additional training or complex processing.
Data Source
AI summary
Disclosed are a CAM-based weakly supervised object localization device and method. The device includes: a feature map extractor configured to extract a feature map of a last convolutional layer in a convolutional neural network (CNN) in a process of applying an image to the CNN; a weight vector binarization unit configured to first binarize a weight vector of a linear layer in a process of sequentially applying the feature map to a pooling layer that generates a feature vector and the linear layer that generates a class label; a feature map binarization unit configured to second binarize the feature map based on the first binarized weight vector; and a class activation map generation unit configured to generate a class activation map for object localization based on the second binarized feature map.