TL;DR
A spatial-channel collaborative modeling framework named SPMixNet is introduced to improve fine-grained 3D scene understanding and 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.
SDA-Reg is introduced, a registration network based on semantically enhanced features and a dual-stream attention architecture that outperforms baseline methods by 1.43%, demonstrating robust accuracy in dynamic environments.
This paper replaces the Stem layer in FRNet with the proposed FD-Stem, which improves feature representation while reducing computational complexity, and introduces long-range modeling capability with limited additional parameters, enabling effective learning of both spatial and channel-wise representations.
The proposed PDES-Net enhances segmentation performance while maintaining advantages in model parameters and inference speed, and achieves a well-balanced trade-off between accuracy and computational efficiency.
Source-Free Domain Adaptation (SFDA) offers a practical solution for deploying LiDAR point cloud segmentation models in new environments where source data cannot be accessed due to privacy constraints and storage limitations. Existing SFDA methods, however, are primarily designed for images and fail to account for the...
The proposed GSSP-KAN is an efficient semantic segmentation network that integrates Kansformer with grouped separable sparse convolution, while the Grouped Separable Sparse Convolution Block (GSSP_Block) reduces redundant self-attention computation and preserves fine-grained local geometric structures.
GFE-Net is rigorously benchmarked on two widely adopted large-scale datasets—S3DIS and SensatUrban—yielding OA/mIoU of 89.6%/73.1% and 93.3%/61.1%, respectively.
We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.
