Walk in the Cloud: Learning Curves for Point Clouds Shape Analysis

Xiang, Tiange; Zhang, Chaoyi; Song, Yang; Yu, Jun; Cai, Weidong

doi:10.1109/iccv48922.2021.00095

Cited by 164 publications

(64 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…VointNet achieves state-of-the-art performance on ShapeNet Core55 retrieval tasks and competitive performance on ModelNet40. Unlike the realistic ScanObjectNN, ModelNet40 is based on CAD 3D models and has saturated in the recent years in terms of performance [17,59,62,70].…”

Section: Resultsmentioning

confidence: 99%

“…Table 4 reports the Category-averaged and Instanceaveraged segmentation mIoU of VointNet compared with recent methods on ShapeNet Parts [66]. We achieve comparative performance to strong point cloud networks [54,62,70]. However, when we compare our VointNet to other multiview segmentation baselines (that work on the Voint space and use the same 2D backbone) like Label Fuse [56] or Mean Fuse [29], our VointNet formulation achieves better performance.…”

Section: D Part Segmentationmentioning

confidence: 95%

“…Baselines. We include PointNet [48], PointNet++ [49], DGCNN [58], KPConv [54], Point Transformer [70] and CurveNet [62] as baselines that use point clouds. We also compare against multi-view classification approaches like MVCNN [53] and MVTN [17] as baselines for classification and retrieval and adopt some of the multi-view segmentation baselines (e.g.…”

Section: Methodsmentioning

confidence: 99%

See 2 more Smart Citations

Voint Cloud: Multi-View Point Cloud Representation for 3D Understanding

Hamdi¹,

Giancola²,

Ghanem³

2021

Preprint

View full text Add to dashboard Cite

Multi-view projection methods have demonstrated promising performance on 3D understanding tasks like 3D classification and segmentation. However, it remains unclear how to combine such multi-view methods with the widely available 3D point clouds. Previous methods use unlearned heuristics to combine features at the point level. To this end, we introduce the concept of the multi-view point cloud (Voint cloud), representing each 3D point as a set of features extracted from several view-points. This novel 3D Voint cloud representation combines the compactness of 3D point cloud representation with the natural view-awareness of multi-view representation. Naturally, we can equip this new representation with convolutional and pooling operations. We deploy a Voint neural network (VointNet) with a theoretically established functional form to learn representations in the Voint space. Our novel representation achieves state-of-the-art performance on 3D classification and retrieval on ScanOb-jectNN, ModelNet40, and ShapeNet Core55. Additionally, we achieve competitive performance for 3D semantic segmentation on ShapeNet Parts. Further analysis shows that VointNet improves the robustness to rotation and occlusion compared to other methods.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: D Part Segmentationmentioning

confidence: 95%

Section: Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Voint Cloud: Multi-View Point Cloud Representation for 3D Understanding

Hamdi¹,

Giancola²,

Ghanem³

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…The most common evaluation is the mean Average Preci-Method IA (%) CA (%) PointTrans. [38] 93.7 90.6 PointCNN [17] 92.2 88.1 PointNet [23] 89.2 86 PointNet++ [24] 90.7 -KPConv [30] 92.9 -SpiderCNN [36] 92.4 -CurveNet [35] 93.8 -GBNet [26] 93.8 91 Ours 92.8 89.6…”

Section: Retrievalmentioning

confidence: 99%

“…Some methods use walks on graphs [22] or on meshes [14]. In [35] it is proposed to use guided walks for point clouds. They generate walks based on a given set of rules and heuristics.…”

Section: Introductionmentioning

confidence: 99%