Related Experiment Video
Updated: Dec 19, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
918
SG-One: Similarity Guidance Network for One-Shot Semantic Segmentation
IEEE Transactions on Cybernetics
|June 5, 2020
Summary
This study introduces SG-One, a novel network for one-shot image semantic segmentation. It effectively segments objects using a single annotated example by leveraging similarity guidance, achieving superior results.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Machine Learning
Background:
- One-shot image semantic segmentation is challenging due to limited supervision.
- Recognizing objects from unseen categories requires robust feature representation.
Purpose of the Study:
- To propose a simple yet effective similarity guidance network (SG-One) for one-shot image semantic segmentation.
- To predict segmentation masks using a single annotated support image.
Main Methods:
- Utilized masked average pooling for robust guidance features from the support image.
- Employed cosine similarity to relate support and query image features.
- Developed a unified, end-to-end trainable framework (SG-One).
Main Results:
- Achieved a mean Intersection over Union (mIoU) score of 46.3% on Pascal VOC 2012.
- Demonstrated superior performance compared to baseline methods.
- Showcased the effectiveness of similarity guidance for segmentation.
Conclusions:
- The proposed SG-One network effectively addresses the one-shot semantic segmentation problem.
- The similarity guidance approach enhances object region recognition with minimal supervision.
- SG-One offers an efficient and unified solution for one-shot segmentation tasks.
