PBP-Net: Point Projection and Back-Projection Network for 3D Point Cloud Segmentation

Yang, Juyoung; Lee, Chanho; Ahn, Pyunghwan; Lee, Haeil; Yi, Eojindl; Kim, Junmo

doi:10.48550/arxiv.2011.00988

“…This series of works usually utilize shared MLP to process each point individually to perform feature extraction. However, their performance is limited since they fail to capture local spatial relationships in the data [45] [44]. Recent approaches begin to concentrate on defining effective convolution kernels for points.…”

Section: Multi-view Based Methodsmentioning

confidence: 99%

Representation Aggregation and Propagation Transformer for 3D Point Cloud Analysis

Zhong,

Yang,

Han

2023

Preprint

0

View full text Add to dashboard Cite

The unstructured property challenges the applications of deep learning methods to understand 3D point clouds. While the recent appearance of Transformer architecture provides an ideal solution to address the aforementioned problem. Therefore, based on Transformer's permutation invariance and capability of building long-range context relationship, in this paper we propose an end-to-end Representation Aggregation and Propagation Transformer (RAPFormer) architecture for 3D point cloud analysis. Specifically, two core components, termed Point-wise Double Transformer and Channel-wise Double Transformer modules, are well designed for explicitly capturing interdependencies among points and channels, respectively. And both models consist of two key operations: aggregation attention mapping the point-wise/channel-wise feature maps into a global space, and propagation attention diffusing the aggregated features back to the input points or channels. We illustrate the effectiveness and competitiveness via extensive quantitative and qualitative experiments on 3D shape classification and segmentation benchmark datasets.

show abstract

“…This series of works usually utilize shared MLP to process each point individually to perform feature extraction. However, their performance is limited since they fail to capture local spatial relationships in the data [21] [23].…”

Section: Deep Learning On Point Cloudsmentioning

confidence: 99%

Point Cloud Learning with Transformer

Han¹,

Kuang²,

Xiao³

2022

Preprint

View full text Add to dashboard Cite

Remarkable performance from Transformer networks in Natural Language Processing promote the development of these models in dealing with computer vision tasks such as image recognition and segmentation. In this paper, we introduce a novel framework, called Multi-level Multi-scale Point Transformer (MLMSPT) that works directly on the irregular point clouds for representation learning. Specifically, a point pyramid transformer is investigated to model features with diverse resolutions or scales we defined, followed by a multi-level transformer module to aggregate contextual information from different levels of each scale and enhance their interactions. While a multi-scale transformer module is designed to capture the dependencies among representations across different scales. Extensive evaluation on public benchmark datasets demonstrate the effectiveness and the competitive performance of our methods on 3D shape classification, segmentation tasks.

show abstract

“…This series of works usually utilize shared MLP to process each point individually to perform feature extraction. However, their performance is limited since they fail to capture local spatial relationships in the data [48] [45]. Recent approaches begin to concentrate on defining effective convolution kernels for points.…”

Section: Deep Learning On Point Cloudsmentioning

confidence: 99%

Point Cloud Learning with Transformer

Han¹,

Kuang²,

Xiao³

2021

Preprint

View full text Add to dashboard Cite

Remarkable performance from Transformer networks in Natural Language Processing promote the development of these models in dealing with computer vision tasks such as image recognition and segmentation. In this paper, we introduce a novel framework, called Multi-level Multi-scale Point Transformer (MLMSPT) that works directly on the irregular point clouds for representation learning. Specifically, a point pyramid transformer is investigated to model features with diverse resolutions or scales we defined, followed by a multi-level transformer module to aggregate contextual information from different levels of each scale and enhance their interactions. While a multi-scale transformer module is designed to capture the dependencies among representations across different scales. Extensive evaluation on public benchmark datasets demonstrate the effectiveness and the competitive performance of our methods on 3D shape classification, part segmentation and semantic segmentation tasks.

show abstract

PBP-Net: Point Projection and Back-Projection Network for 3D Point Cloud Segmentation

Cited by 4 publications

References 0 publications

Representation Aggregation and Propagation Transformer for 3D Point Cloud Analysis

Representation Aggregation and Propagation Transformer for 3D Point Cloud Analysis

Point Cloud Learning with Transformer

Point Cloud Learning with Transformer

Contact Info

Product

Resources

About