To capture local complex shape information and fine-grained features from irregular point cloud, this paper proposes a novel local feature encoder-based network named context-awareness (CA)-Net which is used to solve the challenge of 3D object classification and segmentation. The core of the CA-Net is CA-Encoder, which is based on CA and cross-channel multi-head self-attention (CC-MSA). CA-Encoder uses contextual information awareness to aggregate local features from two levels: point cloud 3D coordinate information and high-dimensional implicitly encoded information, leveraging CC-MSA to learn channel-related information. For different point cloud benchmark tasks, CA-Net uses the local feature enhancement module (classification) and the Up-Transformer (segmentation) which includes a cascaded set of CA-Encoders to solve the problem of feature loss at non-edge points in edge sampling, so that the sampled results can both preserve the shape of the point cloud edges and reconstruct the full internal shape structure of the point cloud. The CA-Net has superior performance in experiments on ModelNet and ShapeNetPart datasets with a classification accuracy of 93.8% and a segmentation accuracy of 85.9%.

With the widespread adoption of 3D data across various scenarios, point cloud analysis remains a significant research topic in the field of 3D computer vision. Most existing studies primarily rely on spatial coordinates and their relative positional relationships to implicitly represent local geometric structures, without fully exploiting the explicit structural information within local regions. Furthermore, the differences between features at various levels in terms of geometric detail and semantic abstraction are also issues worthy of attention. Therefore, we propose a novel representation learning framework, termed DoSF-Net, for point cloud classification and segmentation. The effectiveness of DoSF-Net is primarily based on two core designs: the local structure enhancement operator, Deformable Offset Surface-Relational Convolution (DOSConv), and Scale-Selective Semantic-Geometric Fusion (SSGF). DOSConv consists of two main components: 1) Offset-guided Deformable Center Alignment (ODCA), designed to adaptively adjust local aggregation centers so that the constructed local neighborhoods better align with the actual distribution of local structures; and 2) Surface-Relational Encoding (SRE): aimed at explicitly encoding local surface relationships, more comprehensive characterization of complex local geometric structures. Our proposed Scale-Selective Semantic-Geometric Fusion promotes the complementarity of geometric details and semantic representations through scale-selective cross-layer fusion, mitigating the attenuation of geometric cues during hierarchical propagation. Experimental results on point cloud classification and segmentation datasets validate the effectiveness of our method, with DoSF-Net achieving accuracy rates of 94.1% and 86.4% on the ModelNet40 and ShapeNetPart datasets, respectively.