Related Experiment Video
Updated: Aug 19, 2025
14:38
Creating Objects and Object Categories for Studying Perception and Perceptual Learning
Published on: November 2, 2012
11.9K
Semi-Supervised Instance-Segmentation Model for Feature Transfer Based on Category Attention
Hao Wang1, Juncai Liu1, Changhai Huang2
1School of Computer Science, Sichuan University, Chengdu 610065, China.
Sensors (Basel, Switzerland)
|November 26, 2022
Summary
This study introduces AFT-Mask, a novel semi-supervised instance segmentation model that enhances feature transfer using category attention. The AFT-Mask model improves segmentation accuracy by better utilizing source task characteristics.
Area of Science:
- Computer Vision
- Machine Learning
- Deep Learning
Background:
- Semi-supervised instance segmentation is crucial for accurate object detection.
- Existing transfer learning methods do not fully leverage source task features.
- Pseudo-label generation methods show lower segmentation performance compared to transfer learning.
Purpose of the Study:
- To propose AFT-Mask, an attention-based feature transfer Mask R-CNN model.
- To enhance semi-supervised instance segmentation accuracy.
- To improve the utilization of source task characteristics in transfer learning.
Main Methods:
- Developed a semi-supervised instance segmentation model named AFT-Mask.
- Incorporated category attention using object-classification prediction results.
- Designed a migration-optimization module to connect feature transfer and classification.
Main Results:
- AFT-Mask demonstrated effective knowledge transfer capabilities.
- The model significantly improved the performance of benchmark models in semi-supervised instance segmentation.
- Experimental validation was conducted on two distinct datasets.
Conclusions:
- AFT-Mask offers an effective approach to enhance semi-supervised instance segmentation.
- Category attention improves feature transfer module performance.
- The proposed model advances the state-of-the-art in semi-supervised learning for image segmentation.
