Point Cloud Super Resolution with Adversarial Residual Graph Networks

Wu, Huikai; Zhang, Junge; Huang, Kaiqi

doi:10.48550/arxiv.1908.02111

Cited by 11 publications

(18 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…3D Upsampling. Similar to point cloud completion, several works [51,50,49,21,45] aim at generating dense and uniform point clouds given sparse and non-uniform point sets. PU-Net [51] adopted the PointNet++ [30] as a backbone to extract point features and expand feature dimensions by a series of convolutions.…”

Section: Related Workmentioning

confidence: 99%

“…We adopt an adversarial loss that penalizes inaccurate points from the ground truth to learn the complex point distributions and further improve the performance. Instead of classifying the whole object by predicting a single confidence value like conventional generative adversarial networks (GANs) [21,46], we design a patch-based discriminator to explicitly force every local patch of generated point clouds to have the same pattern with real complete point clouds inspired by [16,45]. We show state-of-the-art quantitative and qualitative results on different datasets by various experiments.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Cascaded Refinement Network for Point Cloud Completion

Wang

Ang

Lee

2020

2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

240

153

View full text Add to dashboard Cite

Point clouds are often sparse and incomplete. Existing shape completion methods are incapable of generating details of objects or learning the complex point distributions. To this end, we propose a cascaded refinement network together with a coarse-to-fine strategy to synthesize the detailed object shapes. Considering the local details of partial input with the global shape information together, we can preserve the existing details in the incomplete point set and generate the missing parts with high fidelity. We also design a patch discriminator that guarantees every local area has the same pattern with the ground truth to learn the complicated point distribution. Quantitative and qualitative experiments on different datasets show that our method achieves superior results compared to existing state-of-the-art approaches on the 3D point cloud completion task. Our source code is available at https: // github.com/ xiaogangw/ cascaded-point-completion.git.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Cascaded Refinement Network for Point Cloud Completion

Wang

Ang

Lee

2020

2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

240

153

View full text Add to dashboard Cite

show abstract

“…Early optimization based point cloud upsampling algorithms resort to shape priors [1,14,24,29]. With the development of utilizing deep networks in dealing with point clouds, some learning based point cloud upsampling methods appeared in recent years [21,31,32,39,42]. Yu et al [44] propose the PU-Net, which is the first deep network for generating a denser and uniform point cloud from a sparser set of points.…”

Section: Point Cloud Upsamplingmentioning

confidence: 99%

“…MPU [43] uses multiscale skip connections to combine local and global features, which are also used in our PCS-Net. Different frameworks such as the generative adversarial network (GAN) [20], the graph convolutional network (GCN) [31,39], and the meta-learning [42] are applied to point set upsampling. Qian et al [32] introduce a geometric-centric neural network, called the PUGeo-Net, to generate new samples around the input points.…”

Section: Point Cloud Upsamplingmentioning

confidence: 99%

Deep Point Cloud Simplification for High-quality Surface Reconstruction

Li¹,

Guo²,

Yang³

et al. 2022

Preprint

View full text Add to dashboard Cite

The growing size of point clouds enlarges consumptions of storage, transmission, and computation of 3D scenes. Raw data is redundant, noisy, and non-uniform. Therefore, simplifying point clouds for achieving compact, clean, and uniform points is becoming increasingly important for 3D vision and graphics tasks. Previous learning based methods aim to generate fewer points for scene understanding, regardless of the quality of surface reconstruction, leading to results with low reconstruction accuracy and bad point distribution. In this paper, we propose a novel point cloud simplification network (PCS-Net) dedicated to high-quality surface mesh reconstruction while maintaining geometric fidelity. We first learn a sampling matrix in a feature-aware simplification module to reduce the number of points. Then we propose a novel double-scale resampling module to refine the positions of the sampled points, to achieve a uniform distribution. To further retain important shape features, an adaptive sampling strategy with a novel saliency loss is designed. With our PCS-Net, the input non-uniform and noisy point cloud can be simplified in a feature-aware manner, i.e., points near salient features are consolidated but still with uniform distribution locally. Experiments demonstrate the effectiveness of our method and show that we outperform previous simplification or reconstruction-oriented upsampling methods.

show abstract

“…The adversarial loss penalizes inaccurate points from the ground truth. Instead of predicting a single classification score like conventional generative adversarial networks (GANs) [17], [28], we adopt a patch-based discriminator to explicitly force every local patch of the generated point clouds to have the same pattern with ground truth point clouds inspired by [29], [30].…”

Section: Introductionmentioning

confidence: 99%

Cascaded Refinement Network for Point Cloud Completion with Self-supervision

Wang¹,

Ang²,

Lee³

2020

Preprint

View full text Add to dashboard Cite

Point clouds are often sparse and incomplete, which imposes difficulties for real-world applications, such as 3D object classification, detection and segmentation. Existing shape completion methods tend to generate coarse shapes of objects without finegrained details. Moreover, current approaches require fully-complete ground truth, which are difficult to obtain in real-world applications. In view of these, we propose a self-supervised object completion method, which optimizes the training procedure solely on the partial input without utilizing the fully-complete ground truth. In order to generate high-quality objects with detailed geometric structures, we propose a cascaded refinement network (CRN) with a coarse-to-fine strategy to synthesize the complete objects. Considering the local details of partial input together with the adversarial training, we are able to learn the complicated distributions of point clouds and generate the object details as realistic as possible. We verify our self-supervised method on both unsupervised and supervised experimental settings and show superior performances. Quantitative and qualitative experiments on different datasets demonstrate that our method achieves more realistic outputs compared to existing state-of-the-art approaches on the 3D point cloud completion task.

show abstract

Point Cloud Super Resolution with Adversarial Residual Graph Networks

Cited by 11 publications

References 34 publications

Cascaded Refinement Network for Point Cloud Completion

Cascaded Refinement Network for Point Cloud Completion

Deep Point Cloud Simplification for High-quality Surface Reconstruction

Cascaded Refinement Network for Point Cloud Completion with Self-supervision

Contact Info

Product

Resources

About