SPNet: Deep 3D Object Classification and Retrieval using Stereographic Projection

Yavartanoo, Mohsen; Kim, Eu Young; Lee, Kyoung Mu

doi:10.48550/arxiv.1811.01571

Cited by 3 publications

(4 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Previous researchers attempt to project a point cloud to 2D images along different directions and apply standard 2D CNN to extract features. The features learned from different images are aggregated to a global feature through a view-pooling layer, then this global feature can be utilized to classify objects [5,14,15]. Although the view based method can achieve high accuracy in the classification task, it is nontrivial to apply this method to segment point cloud, which classifies each point to a specific category.…”

Section: A View Based Methodsmentioning

confidence: 99%

Linked Dynamic Graph CNN: Learning on Point Cloud via Linking Hierarchical Features

Zhang,

Hao,

Wang

et al. 2019

Preprint

View full text Add to dashboard Cite

Learning on point cloud is eagerly in demand because the point cloud is a common type of geometric data and can aid robots to understand environments robustly. However, the point cloud is sparse, unstructured, and unordered, which cannot be recognized accurately by a traditional convolutional neural network (CNN) nor a recurrent neural network (RNN). Fortunately, a graph convolutional neural network (Graph CNN) can process sparse and unordered data. Hence, we propose a linked dynamic graph CNN (LDGCNN) to classify and segment point cloud directly in this paper. We remove the transformation network, link hierarchical features from dynamic graphs, freeze feature extractor, and retrain the classifier to increase the performance of LDGCNN. We explain our network using theoretical analysis and visualization. Through experiments, we show that the proposed LDGCNN achieves state-of-art performance on two standard datasets: ModelNet40 and ShapeNet.

show abstract

Section: A View Based Methodsmentioning

confidence: 99%

Linked Dynamic Graph CNN: Learning on Point Cloud via Linking Hierarchical Features

Zhang,

Hao,

Wang

et al. 2019

Preprint

View full text Add to dashboard Cite

show abstract

“…The advanced tasks are more specific tasks directly solving a practical real world problem. [3] 57.0 RoPS [4] 61.1 FPFH [29] 42.3 SI [26] 59.4 PCA [25] 45.8 SC [27] 61.5 SDF [28] 33.2 LMVCNN [8] 66.1 CGF [7] 65.6 Ours 68.1 SHOT [3] 48.5 RoPS [4] 51.6 FPFH [29] 40.8 SI [26] 55.4 PCA [25] 45.6 SC [27] 58.2 SDF [28] 28.3 Local Descriptors (Deep Learning) LMVCNN [8] 66.1 CGF [7] 65.5 Ours 69.5 Global Descriptors (Deep Learning) RotationNet [34] 98.46 SPNet [32] 97. SPNet [32] 94.2 PANORAMA-ENN [33] 93.28…”

Section: Applicationsmentioning

confidence: 99%

“…These techniques provide global descriptors. It can be seen that the performance of state-of-the-art is greater than 95%, however, such techniques are usually based on alternate representations of point clouds such as projections [32] or ensemble of networks [33] to achieve such results instead of directly processing point clouds.…”

Section: Classificationmentioning

confidence: 99%

DeepPoint3D: Learning discriminative local descriptors using deep metric learning on 3D point clouds

Srivastava

Lall

2019

Pattern Recognition Letters

View full text Add to dashboard Cite

Learning local descriptors is an important problem in computer vision. While there are many techniques for learning local patch descriptors for 2D images, recently efforts have been made for learning local descriptors for 3D points. The recent progress towards solving this problem in 3D leverages the strong feature representation capability of image based convolutional neural networks by utilizing RGB-D or multi-view representations. However, in this paper, we propose to learn 3D local descriptors by directly processing unstructured 3D point clouds without needing any intermediate representation. The method constitutes a deep network for learning permutation invariant representation of 3D points. To learn the local descriptors, we use a multi-margin contrastive loss which discriminates between similar and dissimilar points on a surface while also leveraging the extent of dissimilarity among the negative samples at the time of training. With comprehensive evaluation against strong baselines, we show that the proposed method outperforms state-of-the-art methods for matching points in 3D point clouds. Further, we demonstrate the effectiveness of the proposed method on various applications achieving state-of-the-art results.

show abstract

“…Nevertheless, in the era of deep learning, this representation is o en bypassed because of its irregularity, which does not suit Convolutional Neural Networks (CNNs). Instead, 3D data is o en represented as volumetric grids (Ben-Shabat et al 2018;Maturana and Scherer 2015;Roynard et al 2018;Sedaghat et al 2016b) or multiple 2D projections (Boulch et al 2017;Feng et al 2018a;Kanezaki et al 2018;Su et al 2015;Yavartanoo et al 2018). In some recent works point clouds are utilized and new ways to convolve or pool are proposed (Atzmon et al 2018;Hua et al 2018;omas et al 2019;Xu et al 2018).…”

Section: Introductionmentioning

confidence: 99%

MeshWalker: Deep Mesh Understanding by Random Walks

Lahav,

Tal

2020

Preprint

View full text Add to dashboard Cite

SPNet: Deep 3D Object Classification and Retrieval using Stereographic Projection

Cited by 3 publications

References 0 publications

Linked Dynamic Graph CNN: Learning on Point Cloud via Linking Hierarchical Features

Linked Dynamic Graph CNN: Learning on Point Cloud via Linking Hierarchical Features

DeepPoint3D: Learning discriminative local descriptors using deep metric learning on 3D point clouds

MeshWalker: Deep Mesh Understanding by Random Walks

Contact Info

Product

Resources

About