<p>Cloth-changing person re-identification (CC-ReID) aims to match images of the same person across different camera with significant changes in appearance. The main challenge of CC-ReID is how to keep the model consistently focused on clothing-irrelevant regions. Existing methods primarily focus on hard semantic representation learning to obtain clothing-irrelevant features, such as body shape, silhouette and gait, which are dependent on the characteristics of the specific representation and sensitive to the auxiliary extraction network, causing perception bias in identity-related regions. In contrast, soft biological semantics, which represent inherent identity information embedded in the original image, exhibit greater diversity and stability. To explore the potential of soft biological semantics, we propose a novel Soft Biological Semantic-Guided Explicit–Implicit Learning network (SEILNet), which extracts soft biological semantics at multiple network levels and employs explicit–implicit learning to guide the model in distinguishing identity-relevant from identity-irrelevant information. Specifically, on the one hand, we construct a Multi-level Progressive Feature Enhancement (MPFE) Module and Identity Consistency Module (ICM) to collaboratively guide the model to progressively focus on soft biological semantics across shallow to deep network layers while mitigating the impact of pedestrian pose variations on semantic extraction. On the other hand, we propose Explicit Enhancement Module (EEM) and Implicit Suppression Module (ISM) to further enhance the robustness of the extracted soft biological semantics against noise induced by clothing and other confounding factors. Finally, comparing with current popular methods on three CC-ReID datasets, the experimental results show the superiority of our method. The code is available at <a href="https://github.com/TiAmotws/SEILNet-CCReID">https://github.com/TiAmotws/SEILNet-CCReID</a>.</p>