Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Diverse Representation Embedding for Lifelong Person Re-Identification

Shiben Liu, Huijie Fan, Qiang Wang, Xiai Chen, Zhi Han, Yandong Tang

TL;DR

This work tackles lifelong person re-identification (LReID), where models must preserve previously learned knowledge while adapting to sequential new tasks with limited old-task data and domain shifts. It introduces Diverse Representation Embedding (DRE), a transformer-based framework that generates multiple overlapping representations per instance via multiple class tokens and a Maximum Embedding (ME) mechanism, and uses an Adaptive Constraint Module (ACM) to integrate and separate these representations. It further proposes Knowledge Update (KU) and Knowledge Preservation (KP) strategies to balance adaptive learning for new information and retention of old knowledge through representation- and logit-level mechanisms, including losses such as $L_{id}$, $L_{trip}$, $L_{LLD}$, $L_{con}$, and $LLS$, with an EMA-based slow-update strategy. Extensive experiments across seen and unseen domains, including occluded datasets, demonstrate that DRE outperforms state-of-the-art methods and exhibits strong generalization and robustness in lifelong settings.

Abstract

Lifelong Person Re-Identification (LReID) aims to continuously learn from successive data streams, matching individuals across multiple cameras. The key challenge for LReID is how to effectively preserve old knowledge while incrementally learning new information, which is caused by task-level domain gaps and limited old task datasets. Existing methods based on CNN backbone are insufficient to explore the representation of each instance from different perspectives, limiting model performance on limited old task datasets and new task datasets. Unlike these methods, we propose a Diverse Representations Embedding (DRE) framework that first explores a pure transformer for LReID. The proposed DRE preserves old knowledge while adapting to new information based on instance-level and task-level layout. Concretely, an Adaptive Constraint Module (ACM) is proposed to implement integration and push away operations between multiple overlapping representations generated by transformer-based backbone, obtaining rich and discriminative representations for each instance to improve adaptive ability of LReID. Based on the processed diverse representations, we propose Knowledge Update (KU) and Knowledge Preservation (KP) strategies at the task-level layout by introducing the adjustment model and the learner model. KU strategy enhances the adaptive learning ability of learner models for new information under the adjustment model prior, and KP strategy preserves old knowledge operated by representation-level alignment and logit-level supervision in limited old task datasets while guaranteeing the adaptive learning information capacity of the LReID model. Compared to state-of-the-art methods, our method achieves significantly improved performance in holistic, large-scale, and occluded datasets.

Diverse Representation Embedding for Lifelong Person Re-Identification

TL;DR

This work tackles lifelong person re-identification (LReID), where models must preserve previously learned knowledge while adapting to sequential new tasks with limited old-task data and domain shifts. It introduces Diverse Representation Embedding (DRE), a transformer-based framework that generates multiple overlapping representations per instance via multiple class tokens and a Maximum Embedding (ME) mechanism, and uses an Adaptive Constraint Module (ACM) to integrate and separate these representations. It further proposes Knowledge Update (KU) and Knowledge Preservation (KP) strategies to balance adaptive learning for new information and retention of old knowledge through representation- and logit-level mechanisms, including losses such as , , , , and , with an EMA-based slow-update strategy. Extensive experiments across seen and unseen domains, including occluded datasets, demonstrate that DRE outperforms state-of-the-art methods and exhibits strong generalization and robustness in lifelong settings.

Abstract

Lifelong Person Re-Identification (LReID) aims to continuously learn from successive data streams, matching individuals across multiple cameras. The key challenge for LReID is how to effectively preserve old knowledge while incrementally learning new information, which is caused by task-level domain gaps and limited old task datasets. Existing methods based on CNN backbone are insufficient to explore the representation of each instance from different perspectives, limiting model performance on limited old task datasets and new task datasets. Unlike these methods, we propose a Diverse Representations Embedding (DRE) framework that first explores a pure transformer for LReID. The proposed DRE preserves old knowledge while adapting to new information based on instance-level and task-level layout. Concretely, an Adaptive Constraint Module (ACM) is proposed to implement integration and push away operations between multiple overlapping representations generated by transformer-based backbone, obtaining rich and discriminative representations for each instance to improve adaptive ability of LReID. Based on the processed diverse representations, we propose Knowledge Update (KU) and Knowledge Preservation (KP) strategies at the task-level layout by introducing the adjustment model and the learner model. KU strategy enhances the adaptive learning ability of learner models for new information under the adjustment model prior, and KP strategy preserves old knowledge operated by representation-level alignment and logit-level supervision in limited old task datasets while guaranteeing the adaptive learning information capacity of the LReID model. Compared to state-of-the-art methods, our method achieves significantly improved performance in holistic, large-scale, and occluded datasets.
Paper Structure (16 sections, 17 equations, 3 figures, 7 tables)

This paper contains 16 sections, 17 equations, 3 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Person Re-Identification
  4. Lifelong Person Re-Identification
  5. Diverse representation learning
  6. Proposed Method
  7. Overview
  8. Transformer-based structure
  9. Adaptive constraint module
  10. Knowledge Update
  11. Knowledge Preservation
  12. Experiments
  13. Datasets and Evaluation Metrics
  14. Comparison with State-of-the-Art methods
  15. Ablation Studies
  16. ...and 1 more sections

Figures (3)

  • Figure 1: An illustration of our DRE. Each shape denotes an instance, and color indicates a different representation of the same instance. The blue circle indicates the primary embedding representation after integration. At the top, the generated diverse representation is implemented for integration and push operation. Integrated primary embedding representation (blue circle) presents rich body information. Multiple separate auxiliary embedding representations (orange and purple) maintain discrimination of each instance. At the bottom, based on instance-level, we introduce Knowledge Update (KU) and Knowledge Preservation (KP) strategies to improve model performance in preserving old knowledge and adapting to new information in sequential tasks.
  • Figure 2: An overview of the proposed DRE for LReID. We first explore a pure transformer to LReID and equip it with an adjustment model and a learner model. The adaptive constraint module learns rich and discriminative representations. We then introduce knowledge update and knowledge preservation strategies based on diverse representations that collaboratively achieve knowledge interaction to preserve old knowledge while adapting to new information.
  • Figure 3: Generalization ability on unseen datasets during the training process.