Related Experiment Video
Updated: Jul 26, 2026
03:31
End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
Published on: December 15, 2023
A Global Collaborative Discriminative Denoising Network for Text-to-Image Person Re-Identification
1College of Science, Qingdao University of Technology, Qingdao 266000, China.
Sensors (Basel, Switzerland)
|June 12, 2026
Summary
This study introduces the Global Collaborative Discriminative Denoising Network (GCDD) to improve Text-to-Image Person Re-Identification (TI-ReID). GCDD enhances feature representation, effectively addressing semantic misalignment and noise for more accurate pedestrian retrieval.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Machine Learning
Background:
- Text-to-Image Person Re-Identification (TI-ReID) retrieves pedestrians using text descriptions.
- Current dual-tower models struggle with semantic misalignment and noise from occlusions and clutter.
Purpose of the Study:
- To propose a novel framework, the Global Collaborative Discriminative Denoising Network (GCDD), to enhance TI-ReID performance.
- To mitigate semantic misalignment and noise issues in existing TI-ReID methods.
Main Methods:
- Developed a dual-tower fine-tuning framework using CLIP and BERT encoders.
- Introduced three branches: Discriminative Token Selection (DTS) for filtering, Global-Guided Feature Adaptation (GFA) for recalibration, and Query-Driven Aggregation (QDA) for representation.
- Fused features using a parameter-free averaging strategy.
Main Results:
- GCDD demonstrated strong competitive performance on three standard TI-ReID benchmarks.
- The proposed feature enhancement framework proved effective in improving TI-ReID accuracy.
Conclusions:
- The Global Collaborative Discriminative Denoising Network (GCDD) offers a robust solution for Text-to-Image Person Re-Identification.
- The novel feature enhancement approach effectively tackles challenges like semantic misalignment and noise.
