Generalizable Person Re-identification via Self-Supervised Batch Norm Test-Time Adaption

Abstract

In this paper, we investigate the generalization problem of person re-identification (re-id), whose major challenge is the distribution shift on an unseen domain. As an important tool of regularizing the distribution, batch normalization (BN) has been widely used in existing methods. However, they neglect that BN is severely biased to the training domain and inevitably suffers the performance drop if directly generalized without being updated. To tackle this issue, we propose Batch Norm Test-time Adaption (BNTA), a novel re-id framework that applies the self-supervised strategy to update BN parameters adaptively. Specifically, BNTA quickly explores the domain-aware information within unlabeled target data before inference, and accordingly modulates the feature distribution normalized by BN to adapt to the target domain. This is accomplished by two designed self-supervised auxiliary tasks, namely part positioning and part nearest neighbor matching, which help the model mine the domain-aware information with respect to the structure and identity of body parts, respectively. To demonstrate the effectiveness of our method, we conduct extensive experiments on three re-id datasets and confirm the superior performance to the state-of-the-art methods.

Heterogeneous Test-Time Training for Multi-Modal Person Re-identification

“…This general approach proposed in (Sun et al 2020) is easy to combine with other tasks and effective for them. Many researchers introduce the testtime training to the downstream tasks to improve the generalization of the model on the out-of-distribution test data (Han et al 2022;Wang et al 2020;Shin et al 2022;Liu et al 2022;Gandelsman et al 2022). Shin et al (Shin et al 2022) propose two complementary modules, intra-modal pseudolabel generation, and inter-modal pseudo-label refinement, to take full advantage of self-supervising signals provided by multi-modality.…”

“…Shin et al (Shin et al 2022) propose two complementary modules, intra-modal pseudolabel generation, and inter-modal pseudo-label refinement, to take full advantage of self-supervising signals provided by multi-modality. Han et al (Han et al 2022) propose a test-time training ReID framework to update BN parameters adaptively by two designed self-supervised tasks. However, the self-supervised modules in (Shin et al 2022) are specially designed for point clouds and RGB data.…”

“…However, the self-supervised modules in (Shin et al 2022) are specially designed for point clouds and RGB data. The selfsupervised tasks proposed in (Han et al 2022) do not consider the advantages existing in multi-modal data and introduce additional networks and modules to assist prediction. Therefore, these test-time training strategies cannot be directly applied to multi-modal ReID tasks.…”