Related Experiment Video
Updated: Oct 22, 2025
12:39
A Methodology for Capturing Joint Visual Attention Using Mobile Eye-Trackers
Published on: January 18, 2020
7.9K
TGAN: A simple model update strategy for visual tracking via template-guidance attention network
Kai Yang1, Haijun Zhang1, Dongliang Zhou1
1School of Computer Science and Technology, Harbin Institute of Technology, Shenzhen, 518055, China.
Summary
This study introduces a template-guidance attention network (TGAN) for robust visual tracking. TGAN improves upon existing methods by implicitly updating target templates and refining features, enhancing performance in challenging conditions.
Area of Science:
- Computer Vision
- Machine Learning
- Artificial Intelligence
Background:
- Visual attention is crucial for visual tasks.
- Probabilistic discriminative model prediction (PrDiMP) and Siamese box adaptive network (SiamBAN) are effective visual trackers.
- Existing methods lack online template updating and independent feature vector utilization.
Purpose of the Study:
- To propose a template-guidance attention network (TGAN) for enhanced visual tracking.
- To improve robustness against occlusion and deformation.
- To achieve state-of-the-art results in visual tracking benchmarks.
Main Methods:
- Developed TGAN-I (IoU-Net) and TGAN-S (Siamese) frameworks.
- Implemented implicit template updating and adaptive feature refinement using channel and spatial attention.
- Integrated deformable convolutional networks for improved generalization.
Main Results:
- TGAN-I and TGAN-S demonstrated comprehensive utilization of feature information.
- Achieved state-of-the-art performance on six benchmarks.
- TGAN-I significantly outperformed PrDiMP on VOT2019 and VOT2016 EAO scores.
Conclusions:
- The proposed TGAN frameworks offer a robust solution for visual tracking.
- The implicit template update strategy and attention mechanisms enhance tracking accuracy and resilience.
- TGAN represents a significant advancement in visual tracking technology.
