Related Experiment Video
Updated: Oct 4, 2025
12:49
A Method for 3D Reconstruction and Virtual Reality Analysis of Glial and Neuronal Cells
Published on: September 28, 2019
13.0K
Edge-enhanced instance segmentation by grid regions of interest.
Ying Gao1, Zhiyang Qi1, Dexin Zhao1
1Tianjin Key Laboratory of Intelligence Computing and Novel Software Technology, Tianjin University of Technology, No.391 West Binshui Road, Tianjin, 300384 China.
The Visual Computer
|February 7, 2022
Summary
This study introduces EEMask, an efficient single-stage instance segmentation model. EEMask improves both prediction speed and edge segmentation accuracy for tasks like autonomous driving and medical diagnosis.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Image Processing
Background:
- Instance segmentation is crucial for industrial applications like autonomous driving and medical diagnosis.
- Existing models struggle to balance prediction efficiency with accurate instance edge segmentation.
- Current methods often rely on proposal boxes, which can be computationally intensive.
Purpose of the Study:
- To propose an efficient single-stage instance segmentation model named EEMask (edge-enhanced mask).
- To improve the accuracy of instance edge segmentation without increasing computational overhead.
- To enhance the model's ability to detect and segment individual instances in images.
Main Methods:
- EEMask utilizes grid regions of interest (ROIs) instead of traditional proposal boxes.
- It calculates grid relevance based on distance and grayscale values for ROI generation.
- An edge-enhanced layer is incorporated to improve perception of instance edges.
Main Results:
- EEMask demonstrates a 17.8% speed improvement over BlendMask on average.
- Achieved a mask Average Precision (AP) of 39.9 on the MS COCO dataset.
- Outperformed Mask RCNN by 7.5% and BlendMask by 3.9% in mask AP.
Conclusions:
- EEMask offers an efficient and effective solution for instance segmentation tasks.
- The model achieves state-of-the-art performance in both speed and segmentation accuracy.
- The edge-enhanced layer significantly boosts the model's capability in precise edge detection.
