Related Experiment Video
Updated: Jul 8, 2026
08:25
Combining Eye-tracking Data with an Analysis of Video Content from Free-viewing a Video of a Walk in an Urban Park Environment
Published on: May 7, 2019
9.6K
Unsupervised Visible-Infrared ReID via Pseudo-Label Correction and Modality-Level Alignment.
IEEE Transactions on Neural Networks and Learning Systems
|September 9, 2025
Summary
This study introduces a new framework for unsupervised visible-infrared person reidentification (UVI-ReID), improving accuracy by correcting noisy labels and aligning cross-modality features. The PRAISE method enhances human detection in diverse environments without manual labeling.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Machine Learning
Background:
- Unsupervised visible-infrared person reidentification (UVI-ReID) is crucial for human detection across different environments without requiring labeled data.
- Existing UVI-ReID methods face challenges with noisy pseudo-labels from clustering and potential misalignment of features across visible and infrared modalities.
- Theoretical analysis introduced an interpretable generalization upper bound to guide the development of improved UVI-ReID frameworks.
Purpose of the Study:
- To propose a novel unsupervised cross-modality person reidentification framework (PRAISE) addressing key challenges in UVI-ReID.
- To enhance the accuracy and robustness of person reidentification systems using both visible and infrared imagery.
- To reduce the modality gap and learn identity-discriminative, modality-invariant features.
Main Methods:
- Developed a pseudo-label correction (PLC) strategy using a beta mixture model (BMM) to rectify misclustered labels and a perceptual term in contrastive learning.
- Introduced a modality-level alignment (MLA) strategy to generate paired visible-infrared latent features and align their labeling functions.
- Implemented a novel unsupervised cross-modality person reidentification framework (PRAISE).
Main Results:
- The proposed PRAISE framework achieved state-of-the-art (SOTA) performance on two benchmark datasets for unsupervised visible-ReID.
- The PLC strategy effectively mitigated issues arising from noisy pseudo-labels in the clustering process.
- The MLA strategy successfully reduced the modality gap, leading to more discriminative and invariant features.
Conclusions:
- The PRAISE framework offers a significant advancement in unsupervised visible-infrared person reidentification.
- The combination of PLC and MLA strategies effectively addresses the limitations of previous UVI-ReID methods.
- This research contributes to more robust and accurate human detection systems in complex, multi-modal environments.
Related Concept Videos
01:09Infrared (IR) Spectroscopy: Overview
When electromagnetic radiation passes through a material, atoms or molecules transition from a lower to a higher energy state by absorbing radiation corresponding to the energy difference between the two states. The absorption of infrared (IR) radiation causes transitions between vibrational energy levels in a molecule. Therefore, IR spectroscopy is a useful analytical tool for determining the molecular structure of molecules.
Different compounds display unique properties due to their...
Different compounds display unique properties due to their...
01:24IR Frequency Region: X–H Stretching
In IR spectroscopy, signals produced by the X−H bonds (such as C−H, O−H, or N−H) can be observed in the frequency range of 2700–4000 cm–1. The C−H stretching vibration forms sharp bands in the region 2850–3000 cm–1. The presence of the O−H stretching vibration leads to the forming of an absorption band in the frequency range 3650–3200 cm−1. At the same time, N−H stretching can be confirmed by absorption bands in the 3500–3100 cm−1 range. Even though both O−H and N−H bonds vibrate at a similar...