Update on Literatures

[hansen7]

Point Cloud Completion

• "Topnet: Structural point cloud decoder", CVPR 2019
• "3D Shape Completion with Multi-view Consistent Inference", AAAI 2020
• "Morphing and Sampling Network for Dense Point Cloud Completion", AAAI 2020
• "Cascaded Refinement Network for Point Cloud Completion", CVPR 2020
• "PF-Net: Point Fractal Network for 3D Point Cloud Completion", CVPR 2020
• "Point Cloud Completion by Skip-Attention Network With Hierarchical Folding", CVPR 2020
• "GRNet: Gridding Residual Network for Dense Point Cloud Completion", ECCV 2020
• "SoftPoolNet: Shape Descriptor for Point Cloud Completion and Classification", ECCV 2020
• "Weakly-supervised 3D Shape Completion in the Wild", ECCV 2020
• "Variational Relational Point Completion Network", CVPR 2021

[hansen7]

Unsupervised/Representation Learning on Point Cloud

• "Unsupervised Learning of Shape and Pose with Differentiable Point Clouds", NeurIPS 2018
• "Self-Supervised Deep Learning on Point Clouds by Reconstructing Space", NeurIPS 2019
• "Unsupervised Learning of Intrinsic Structural Representation Points", CVPR 2020 No need to compare, they does not outperform ''3D jigsaw'' with the same configs
• "Unsupervised Deep Shape Descriptor with Point Distribution Learning", CVPR 2020 Not Comparable, focus on different tasks (shape correspondence).
• "PointContrast: Unsupervised Pre-training for 3D Point Cloud Understanding", ECCV 2020 Not Comparable, they develop a new baseline model and mainly compare pre-trained and random initialisation on their own baselines, and mostly focus on the semantic segmentation, has been used in the follow-up:
• • "Exploring Data-Efficient 3D Scene Understanding with Contrastive Scene Contexts", CVPR'21 sub
• "Unsupervised Detection of Distinctive Regions on 3D Shapes", ACM ToG 2020 Interesting paper, worth a cite, results are not directly comparable to ours
• "PDE-Inspired Algorithms for Semi-Supervised Learning on Point Clouds", arXiv 2019 from peers
• "Self-Supervised Few-Shot Learning on Point Clouds", NeurIPS 2020 we also conduct and outperform this in fsl setting.
• "CaSPR: Learning Canonical Spatiotemporal Point Cloud Representations", NeurIPS 2020 also use reconstruction as the training task

[hansen7]

Our NRS Paper: https://arxiv.org/abs/1911.07845

[ShengyuH]

Thanks for your summery!

For self-supervised session, another two follow-ups for PointContrast:

• P4contrast: Contrastive learning with pairs of point-pixel pairs for rgb-d scene understanding, CVPR2021 sub
• Self-supervisedpretraining of 3d features on any point-cloud, CVPR2021 sub

[hansen7]

Thanks @ShengyuH , I've read both but am a bit lazy to update。。。

It turns out there are not much breakthroughs in the contrastive learning for point cloud models, to be honest (i.e., see Matthias and Alexei's discussion on the completion and contrastive pre-training for 3D, ~1:07:00 in this video). But the augmentation designs in the DepthContrast paper is a bit novel, w.r.t this discussion on what should not be contrastive in images contrastive learning

Feel free if you have anything in mind and happy to collaborate :)