.png){kind=icon}
.png){kind=icon}
Return
Spmixnet: spatial-channel collaborative modeling for enhanced small-object segmentation in LiDAR point clouds
.png){kind=icon}
PRE
AI
H
C
Y
G
X
DOI：10.1007/s00371-026-04664-y.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Point cloud semantic segmentation supports critical applications including autonomous driving and smart cities, yet performance remains limited on small or slender objects due to spatial discontinuity and unstable channel-wise semantic representations. This work introduces a spatial-channel collaborative modeling framework named SPMixNet to improve fine-grained 3D scene understanding. A grid-aware spatial mixing module (GASM) enhances local spatial continuity during structured projection, while a hierarchical group channel mixing module (HGCM) stabilizes semantic feature representation across channels. Experiments on the SemanticKITTI and Rellis-3D datasets demonstrate the effectiveness of the proposed method. Spmixnet achieves mIoU scores of 68.7% and 38.16% on SemanticKITTI and Rellis-3D, respectively, while delivering competitive performance on several small-object categories. The proposed design effectively mitigates the structural and representational limitations of existing projection-based methods. The code is available at https://github.com/SunChengGong/SPMixNet . The proposed design effectively mitigates the structural and representational limitations of existing projection-based methods, providing a promising solution for accurate segmentation of small and structurally complex objects in large-scale 3D scenes.
Keywords:
Computer vision
3D semantic segmentation
Small-object segmentation
Spatial-channel modeling
LiDAR point clouds
Autonomous driving
Journal
IF:
2.9
Papers:
4.5K
Citations:
6.5K
Organization
Cited Papers