Related Experiment Video
Updated: Jul 16, 2026
12:39
A Methodology for Capturing Joint Visual Attention Using Mobile Eye-Trackers
Published on: January 18, 2020
7.6K
PPBI: Pose-Guided Partial-Attention Network with Batch Information for Occluded Person Re-Identification.
Jianhai Cui1, Yiping Chen2, Binbin Deng3
1School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China.
Sensors (Basel, Switzerland)
|February 13, 2025
Summary
This study introduces a pose-guided partial-attention network with batch information (PPBI) to improve occluded person re-identification (ReID) by refining key-point relationships and modeling inter-image interactions. The PPBI framework significantly enhances performance on challenging ReID tasks.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Machine Learning
Background:
- Occluded person re-identification (ReID) is challenging due to missing appearance data caused by occlusions.
- Existing ReID methods often overlook relational information within image batches, limiting performance.
Purpose of the Study:
- To develop a novel framework, the pose-guided partial-attention network with batch information (PPBI), to enhance spatial and relational learning for occluded ReID.
- To address occlusion-induced inconsistencies and improve feature extraction in person ReID.
Main Methods:
- Proposing a node optimization network (NON) to refine pedestrian key-point relationships and mitigate occlusion effects.
- Introducing a key-point batch attention (KBA) module to model inter-image interactions within batches.
- Implementing a correction of hard mining (CHM) and batch enhancement (BE) modules to handle misclassifications and strengthen attention.
Main Results:
- The PPBI framework demonstrated robust performance on both occluded and holistic ReID tasks.
- Achieved a 2.7% mAP improvement over the HoNeT method on the Occluded-Duke dataset.
Conclusions:
- The proposed PPBI framework effectively enhances person ReID in the presence of occlusions by leveraging spatial and relational learning.
- PPBI's novel components, NON and KBA, significantly contribute to mitigating occlusion challenges and improving ReID accuracy.
