The paper introduces a novel framework called Robust Pseudo-label Learning with Neighbor Relation (RPNR) for unsupervised visible-infrared person re-identification (USVI-ReID). The main challenges in USVI-ReID are obtaining robust pseudo-labels and establishing reliable cross-modality correspondences. To address these issues, RPNR includes several key modules: 1. **Noisy Pseudo-label Calibration (NPC)**: This module corrects noisy pseudo-labels by calibrating them using reliable neighbor samples, reducing the impact of intra-class variations. 2. **Neighbor Relation Learning (NRL)**: This module reduces intra-class variations by modeling potential interactions between all samples, promoting the model to cluster closely with neighbor samples. 3. **Optimal Transport Prototype Matching (OTPM)**: This module establishes reliable cross-modality correspondences by using optimal transport to align visible and infrared prototypes. 4. **Memory Hybrid Learning (MHL)**: This module learns both modality-specific and modality-invariant information by blending two modality-specific memories, effectively bridging the significant cross-modality gaps. The proposed method is evaluated on two popular benchmarks, SYSU-MM01 and RegDB, demonstrating superior performance compared to existing state-of-the-art methods. The experiments show that RPNR outperforms current methods by an average Rank-1 improvement of 10.3%. The source code will be released soon.The paper introduces a novel framework called Robust Pseudo-label Learning with Neighbor Relation (RPNR) for unsupervised visible-infrared person re-identification (USVI-ReID). The main challenges in USVI-ReID are obtaining robust pseudo-labels and establishing reliable cross-modality correspondences. To address these issues, RPNR includes several key modules: 1. **Noisy Pseudo-label Calibration (NPC)**: This module corrects noisy pseudo-labels by calibrating them using reliable neighbor samples, reducing the impact of intra-class variations. 2. **Neighbor Relation Learning (NRL)**: This module reduces intra-class variations by modeling potential interactions between all samples, promoting the model to cluster closely with neighbor samples. 3. **Optimal Transport Prototype Matching (OTPM)**: This module establishes reliable cross-modality correspondences by using optimal transport to align visible and infrared prototypes. 4. **Memory Hybrid Learning (MHL)**: This module learns both modality-specific and modality-invariant information by blending two modality-specific memories, effectively bridging the significant cross-modality gaps. The proposed method is evaluated on two popular benchmarks, SYSU-MM01 and RegDB, demonstrating superior performance compared to existing state-of-the-art methods. The experiments show that RPNR outperforms current methods by an average Rank-1 improvement of 10.3%. The source code will be released soon.