MF-MOS: A Motion-Focused Model for Moving Object Segmentation Moving object segmentation (MOS) is crucial for autonomous driving systems to accurately perceive dynamic objects such as pedestrians and vehicles. This paper introduces MF-MOS, a novel motion-focused model for LiDAR-based MOS. The model uses a dual-branch structure, with one branch focusing on motion information from residual maps and the other on semantic features from range images. The model's core idea is to decouple spatial-temporal information by capturing motion from residual maps and generating semantic features from range images, which are used as guidance for the motion branch. This approach effectively utilizes both range images and residual maps, significantly improving the performance of the LiDAR-based MOS task. MF-MOS achieved a leading IoU of 76.7% on the SemanticKITTI dataset, demonstrating state-of-the-art performance. The model also includes a 3D Spatial-Guided Information Enhancement Module (SIEM) to provide additional spatial guidance and a distribution-based data augmentation method to improve network robustness. Extensive experiments show that MF-MOS outperforms existing methods in both validation and test sets, achieving the highest accuracy on the SemanticKITTI-MOS dataset. The model is efficient and can be applied to other range-based methods.MF-MOS: A Motion-Focused Model for Moving Object Segmentation Moving object segmentation (MOS) is crucial for autonomous driving systems to accurately perceive dynamic objects such as pedestrians and vehicles. This paper introduces MF-MOS, a novel motion-focused model for LiDAR-based MOS. The model uses a dual-branch structure, with one branch focusing on motion information from residual maps and the other on semantic features from range images. The model's core idea is to decouple spatial-temporal information by capturing motion from residual maps and generating semantic features from range images, which are used as guidance for the motion branch. This approach effectively utilizes both range images and residual maps, significantly improving the performance of the LiDAR-based MOS task. MF-MOS achieved a leading IoU of 76.7% on the SemanticKITTI dataset, demonstrating state-of-the-art performance. The model also includes a 3D Spatial-Guided Information Enhancement Module (SIEM) to provide additional spatial guidance and a distribution-based data augmentation method to improve network robustness. Extensive experiments show that MF-MOS outperforms existing methods in both validation and test sets, achieving the highest accuracy on the SemanticKITTI-MOS dataset. The model is efficient and can be applied to other range-based methods.