Grid-Based Image Segmentation for Faster, Finer Instance Masks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing instance segmentation methods are time-consuming and require a large amount of memory, with inadequate segmentation granularity.
Innovation Solution
An image segmentation method that involves obtaining a foreground target box, dividing the region of interest into grids, predicting features for each grid, and combining this with semantic features to achieve instance segmentation, while reducing the output size to 1×S², thereby improving segmentation granularity and reducing memory and time requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing instance segmentation methods are used, then segmentation can be performed, but the processing time is excessive and memory consumption is high
Solution Approach 1:
The patent divides the region of interest into multiple grids and processes each grid independently to predict target existence. This segmentation approach parallelizes the computation, reducing overall processing time while maintaining segmentation accuracy.
Solution Approach 2:
The patent extracts only the necessary features for instance segmentation by focusing on grid-based target existence prediction rather than processing entire feature maps. This extraction of essential information reduces computational overhead and memory requirements.
2Quantity of substance
If existing instance segmentation methods are used, then segmentation can be achieved, but a large amount of memory is required
Solution Approach 1:
By segmenting the region of interest into grids and processing each grid separately, the patent reduces the memory footprint compared to processing the entire feature map at once. Each grid requires only a portion of the total memory resources.
Solution Approach 2:
The patent changes the output representation from a full feature map to a grid-based probability map with dimensions 1×S². This dimensional transformation significantly reduces memory consumption while preserving the essential segmentation information.
3Measurement precision
If existing instance segmentation methods are used, then segmentation can be performed, but the granularity cannot reach the semantic level
Solution Approach 1:
The patent applies local quality by making different parts of the region of interest (each grid) have different processing characteristics. Each grid's target existence probability is predicted independently, allowing for fine-grained semantic-level segmentation precision.
Solution Approach 2:
The patent uses the grid-based prediction results to construct the final instance segmentation mask by copying and combining the predicted target existence probabilities. This copying approach achieves high granularity without requiring complex post-processing operations.
Data Source
AI summary
Disclosed are an image segmentation method, a training method for an image segmentation network, an electronic equipment, and a storage medium. The method includes: sending an input image to an image segmentation network; obtaining a first foreground target box of the input image; obtaining a first region of interest and a first region-of-interest feature map of the input image based on the first foreground target box; dividing the first region of interest into grids, predicting a corresponding feature of each grid in the first region of interest based on the first region-of-interest feature map, obtaining a semantic feature of each pixel in the first region of interest; and obtaining an instance segmentation result based on the corresponding feature of each grid in the first region of interest, information of the first foreground target box, and the semantic feature of each pixel in the first region of interest.