Instance Segmentation via Attention Feature Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image segmentation technologies, particularly in instance segmentation, face challenges in distinguishing between different instances of the same category in images, which is crucial for applications like portrait photography and AR, where precise segmentation is necessary.
Innovation Solution
The method involves pre-training a segmentation network to generate feature fusion maps that weight pixel values based on attention feature maps, allowing for precise identification and differentiation of instance pixels by considering pixel weights, and optionally using bounding boxes for instance identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If instance segmentation is performed to distinguish different instances of the same category, then segmentation precision is improved, but computational complexity and processing time increase
Solution Approach 1:
The segmentation network is divided into multiple functional modules including feature extraction modules, instance classification modules, and pixel-level segmentation modules. Each module handles specific aspects of the segmentation task, allowing the system to achieve high precision while managing computational complexity through modular processing
Solution Approach 2:
The network performs preliminary feature extraction and instance detection before detailed pixel-level segmentation. By pre-identifying instance boundaries and characteristics in earlier processing stages, the system reduces the computational burden on subsequent segmentation stages while maintaining high precision
2Measurement precision
If attention feature maps with different pixel values are used to weight pixels, then instance pixel differentiation accuracy is improved, but computational resources and processing time increase
Solution Approach 1:
The attention feature maps apply different weighting values to different regions and pixels within the image based on their importance for instance segmentation. By focusing computational attention on critical pixels and regions while using lighter weighting for less important areas, the system achieves high pixel differentiation accuracy without uniformly processing all pixels at maximum computational cost
Solution Approach 2:
The system dynamically adjusts attention weights and pixel value parameters based on the input image characteristics and instance properties. By changing these parameters adaptively rather than using fixed values, the network optimizes processing efficiency for different scenarios while maintaining high segmentation accuracy
Data Source
AI summary
An instance segmentation method and apparatus are provided. A to-be-trained segmentation network performs the following processing on each instance group that is in a sample original image and that is of pixels of a marked instance, where each instance group includes at least one marked instance, and the processing includes: predicting at least two different first basic feature maps and a first attention feature map corresponding to each first basic feature map; performing weighted processing on the at least two first basic feature maps and pixel values of respective first attention feature maps corresponding to the at least two first basic feature maps, to obtain a first feature fusion map; and training the to-be-trained segmentation network based on the first feature fusion map and the sample original image. A segmentation model can precisely determine pixels of an instance in an original image.