Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Lifelong Person Re-Identification with Backward-Compatibility

Minyoung Oh, Jae-Young Sim

TL;DR

This paper attempts to train the model using the continuously incoming datasets while maintaining the model’s compatibility toward the previously trained old models without re-computing the features of the old gallery images, and designs the cross-model compatibility loss based on the contrastive learning with respect to the replay features.

Abstract

Lifelong person re-identification (LReID) assumes a practical scenario where the model is sequentially trained on continuously incoming datasets while alleviating the catastrophic forgetting in the old datasets. However, not only the training datasets but also the gallery images are incrementally accumulated, that requires a huge amount of computational complexity and storage space to extract the features at the inference phase. In this paper, we address the above mentioned problem by incorporating the backward-compatibility to LReID for the first time. We train the model using the continuously incoming datasets while maintaining the model's compatibility toward the previously trained old models without re-computing the features of the old gallery images. To this end, we devise the cross-model compatibility loss based on the contrastive learning with respect to the replay features across all the old datasets. Moreover, we also develop the knowledge consolidation method based on the part classification to learn the shared representation across different datasets for the backward-compatibility. We suggest a more practical methodology for performance evaluation as well where all the gallery and query images are considered together. Experimental results demonstrate that the proposed method achieves a significantly higher performance of the backward-compatibility compared with the existing methods. It is a promising tool for more practical scenarios of LReID.

Lifelong Person Re-Identification with Backward-Compatibility

TL;DR

This paper attempts to train the model using the continuously incoming datasets while maintaining the model’s compatibility toward the previously trained old models without re-computing the features of the old gallery images, and designs the cross-model compatibility loss based on the contrastive learning with respect to the replay features.

Abstract

Lifelong person re-identification (LReID) assumes a practical scenario where the model is sequentially trained on continuously incoming datasets while alleviating the catastrophic forgetting in the old datasets. However, not only the training datasets but also the gallery images are incrementally accumulated, that requires a huge amount of computational complexity and storage space to extract the features at the inference phase. In this paper, we address the above mentioned problem by incorporating the backward-compatibility to LReID for the first time. We train the model using the continuously incoming datasets while maintaining the model's compatibility toward the previously trained old models without re-computing the features of the old gallery images. To this end, we devise the cross-model compatibility loss based on the contrastive learning with respect to the replay features across all the old datasets. Moreover, we also develop the knowledge consolidation method based on the part classification to learn the shared representation across different datasets for the backward-compatibility. We suggest a more practical methodology for performance evaluation as well where all the gallery and query images are considered together. Experimental results demonstrate that the proposed method achieves a significantly higher performance of the backward-compatibility compared with the existing methods. It is a promising tool for more practical scenarios of LReID.
Paper Structure (20 sections, 8 equations, 5 figures, 7 tables)

This paper contains 20 sections, 8 equations, 5 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Person Re-Identification
  4. Lifelong Person Re-Identification
  5. Backward-Compatible Training
  6. Methodology
  7. Problem Formulation
  8. Baseline Network Training
  9. Cross-Model Compatibility Loss
  10. Part-Assisted Knowledge Consolidation
  11. Experimental Results
  12. Experimental Setup
  13. Datasets.
  14. Evaluation Protocol.
  15. Implementation Details.
  16. ...and 5 more sections

Figures (5)

  • Figure 1: The concept of the proposed lifelong person re-identification with backward-compatibility. (a) Conventional training. (b) Lifelong learning. (c) Backward-compatible training. (d) Lifelong learning with the backward-compatibility.
  • Figure 2: Illustration of the training and inference procedures in the proposed framework.
  • Figure 3: Part-assisted knowledge consolidation.
  • Figure 4: Visualization of the Top-10 matching results of the proposed method compared with that of the existing methods, where all gallery features of $\boldsymbol{\mathcal{F}}_\mathrm{G}$ and all query features of $\boldsymbol{\mathcal{F}}_\mathrm{Q}$ are considered simultaneously. The true and false matching results are respectively depicted in the green and red colors, respectively. MA: Market1501, DU: DukeMTMC, MS: MSMT17.
  • Figure 5: Visualization of the feature distributions on Market1501. Different colors indicate different identities and different shapes represent the features extracted from different models, respectively.