Vicinity-Based Abstraction: VA-DGCNN Architecture for Noisy 3D Indoor Object Classification

Walczak, Jakub; Wojciechowski, Adam; Najgebauer, Patryk; Scherer, Rafał

doi:10.1007/978-3-030-77967-2_19

Cited by 1 publication

(2 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Shortcuts were added between the different layers to link the hierarchical features to calculate useful edge vectors. VA-DGCNN [33] proposed a novel, feature-preserving vicinity abstraction (VA) layer for the EdgeConv module. Unlike the original DGCNN, local information is aggregated before further processing, rather than processed one point at a time with neighbors.…”

Section: Graph Convolutional Networkmentioning

confidence: 99%

“…Therefore, evaluating area 5 can well test the generalization performance of the network. We compare the classical semantic segmentation methods [5,8,16,28,45], the semantic and instance joint segmentation methods [25,26], the GCN based semantic segmentation methods [17,18,29,33], and the results are shown in Table 1. [17] 49.0 -83.2 91.1 97.3 74.5 0.0 11.9 49.5 33.5 66.9 69.4 20.5 47.5 34.7 40.8 3D-GCN [18] 51.9 -84.6 91.4 97.1 75.9 0.1 22.3 43.5 30.1 71.5 79.4 21.9 53.7 42.9 44.9 ASIS [25] 53.4 60.9 86.9 92.0 98.0 75.3 0.0 10.1 49.9 24.2 72.9 78.1 33.4 58.4 51.0 50.7 JSNet++ [26] 58.0 --93.7 98.5 80.5 0.0 16.9 57.2 41.9 76.8 84.7 30.5 60.2 58.3 54.9 RandLA-Net [28] We notice that all methods perform similarly in the ceiling, floor, and beam categories, as they can be easily classified based on the location in the room.…”

Section: S3dis Datasetmentioning

confidence: 99%

See 1 more Smart Citation

SGT-Net: A Transformer-Based Stratified Graph Convolutional Network for 3D Point Cloud Semantic Segmentation

Liu,

Chi,

et al. 2024

CMC

View full text Add to dashboard Cite

In recent years, semantic segmentation on 3D point cloud data has attracted much attention. Unlike 2D images where pixels distribute regularly in the image domain, 3D point clouds in non-Euclidean space are irregular and inherently sparse. Therefore, it is very difficult to extract long-range contexts and effectively aggregate local features for semantic segmentation in 3D point cloud space. Most current methods either focus on local feature aggregation or long-range context dependency, but fail to directly establish a global-local feature extractor to complete the point cloud semantic segmentation tasks. In this paper, we propose a Transformer-based stratified graph convolutional network (SGT-Net), which enlarges the effective receptive field and builds direct long-range dependency. Specifically, we first propose a novel dense-sparse sampling strategy that provides dense local vertices and sparse long-distance vertices for subsequent graph convolutional network (GCN). Secondly, we propose a multi-key self-attention mechanism based on the Transformer to further weight augmentation for crucial neighboring relationships and enlarge the effective receptive field. In addition, to further improve the efficiency of the network, we propose a similarity measurement module to determine whether the neighborhood near the center point is effective. We demonstrate the validity and superiority of our method on the S3DIS and ShapeNet datasets. Through ablation experiments and segmentation visualization, we verify that the SGT model can improve the performance of the point cloud semantic segmentation.

show abstract

Section: Graph Convolutional Networkmentioning

confidence: 99%