Related Experiment Video
Updated: Jun 27, 2026
07:34
Utilizing vmTracking to Improve the Accuracy of Multi-Animal Pose Estimation in Rodent Social Behavior Studies
Published on: November 7, 2025
TrackRefine: A Plug-and-Play Decoupled Enhancement Framework for Online Multi-Object Tracking and Segmentation.
Longfei Qie1, Chunlei Chai1, Ruixue Wang1
1School of Mechanical and Electrical Engineering, Beijing University of Chemical Technology, Beijing 100029, China.
Sensors (Basel, Switzerland)
|June 26, 2026
Summary
TrackRefine enhances online multi-object tracking and segmentation (MOTS) by refining mask quality and improving identity association, even during occlusions. This plug-and-play framework offers a practical solution for complex tracking scenarios.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Machine Learning
Background:
- Existing online decoupled multi-object tracking and segmentation (MOTS) methods struggle with mask quality, memory corruption during occlusion, and unstable trajectory recovery.
- These limitations hinder performance in complex real-world scenarios.
Purpose of the Study:
- To propose TrackRefine, a novel plug-and-play decoupled enhancement framework for online MOTS.
- To improve mask quality, identity modeling, and trajectory recovery in challenging tracking environments.
Main Methods:
- Introduced a lightweight Fast GrabCut-based module for mask refinement.
- Developed a multimodal long-short-term memory bank integrating appearance, semantic, and shape cues for robust identity modeling.
- Implemented a progressive three-stage association strategy for stable matching and long-term trajectory recovery.
Main Results:
- Achieved state-of-the-art performance on MOTS20 (69.4 sMOTSA, 82.7 MOTSA).
- Demonstrated strong results on KITTI MOTS for both pedestrians and cars.
- Verified plug-and-play flexibility and effectiveness across different front-end segmenters.
Conclusions:
- TrackRefine offers an efficient and practical solution for online MOTS in complex scenarios.
- The decoupled enhancement framework significantly improves tracking performance without altering existing architectures.
