Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
102 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Score-Based Point Cloud Denoising (2107.10981v5)

Published 23 Jul 2021 in cs.CV

Abstract: Point clouds acquired from scanning devices are often perturbed by noise, which affects downstream tasks such as surface reconstruction and analysis. The distribution of a noisy point cloud can be viewed as the distribution of a set of noise-free samples $p(x)$ convolved with some noise model $n$, leading to $(p * n)(x)$ whose mode is the underlying clean surface. To denoise a noisy point cloud, we propose to increase the log-likelihood of each point from $p * n$ via gradient ascent -- iteratively updating each point's position. Since $p * n$ is unknown at test-time, and we only need the score (i.e., the gradient of the log-probability function) to perform gradient ascent, we propose a neural network architecture to estimate the score of $p * n$ given only noisy point clouds as input. We derive objective functions for training the network and develop a denoising algorithm leveraging on the estimated scores. Experiments demonstrate that the proposed model outperforms state-of-the-art methods under a variety of noise models, and shows the potential to be applied in other tasks such as point cloud upsampling. The code is available at \url{https://github.com/luost26/score-denoise}.

View on arXiv
Definition Search Book Streamline Icon: https://streamlinehq.com
References (37)
  1. Point set surfaces. In Proceedings Visualization, 2001. VIS’01., pages 21–29. IEEE, 2001.
  2. ℓℓ\ellroman_ℓ1-sparse reconstruction of sharp point set surfaces. ACM Transactions on Graphics (TOG), 29(5):1–12, 2010.
  3. Learning gradient fields for shape generation. In Proceedings of the European Conference on Computer Vision (ECCV), 2020.
  4. Estimating differential quantities using polynomial fitting of osculating jets. Computer Aided Geometric Design, 22(2):121–146, 2005.
  5. Shapenet: An information-rich 3d model repository. arXiv preprint arXiv:1512.03012, 2015.
  6. The bilateral filter for point clouds. Image Processing On Line, 7:278–287, 2017.
  7. 3d point cloud denoising via deep neural network based local surface estimation. In ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 8553–8557. IEEE, 2019.
  8. A point set generation network for 3d object reconstruction from a single image. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 605–613, 2017.
  9. Bilateral mesh denoising. In ACM SIGGRAPH 2003 Papers, pages 950–953. 2003.
  10. Blensor: Blender sensor simulation toolbox. In International Symposium on Visual Computing, pages 199–208. Springer, 2011.
  11. Total denoising: Unsupervised learning of 3d point cloud cleaning. In Proceedings of the IEEE International Conference on Computer Vision, pages 52–60, 2019.
  12. Feature graph learning for 3D point cloud denoising. IEEE Transactions on Signal Processing, 68:2841–2856, 2020.
  13. Dynamic point cloud denoising via manifold-to-manifold distance. arXiv preprint arXiv:2003.08355, 2020.
  14. Graph Signal Processing for geometric data and beyond: Theory and applications. arXiv preprint arXiv:2008.01918, 2020.
  15. Densely connected convolutional networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4700–4708, 2017.
  16. Edge-aware point set resampling. ACM transactions on graphics (TOG), 32(1):1–12, 2013.
  17. Aapo Hyvärinen. Estimation of non-normalized statistical models by score matching. Journal of Machine Learning Research, 6(Apr):695–709, 2005.
  18. A tutorial on energy-based learning. Predicting structured data, 1(0), 2006.
  19. Pu-gan: a point cloud upsampling adversarial network. In Proceedings of the IEEE International Conference on Computer Vision, pages 7203–7212, 2019.
  20. Densepoint: Learning densely contextual representation for efficient point cloud processing. In Proceedings of the IEEE International Conference on Computer Vision, pages 5239–5248, 2019.
  21. Differentiable manifold reconstruction for point cloud denoising. In Proceedings of the 28th ACM International Conference on Multimedia, pages 1330–1338, 2020.
  22. Point cloud denoising via moving rpca. In Computer Graphics Forum, volume 36, pages 123–137. Wiley Online Library, 2017.
  23. Learning graph-convolutional representations for point cloud denoising. arXiv preprint arXiv:2007.02578, 2020.
  24. Pointnet: Deep learning on point sets for 3d classification and segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 652–660, 2017.
  25. Pointnet++: Deep hierarchical feature learning on point sets in a metric space. In Advances in neural information processing systems, pages 5099–5108, 2017.
  26. Pointcleannet: Learning to denoise and remove outliers from dense point clouds. In Computer Graphics Forum, volume 39, pages 185–203. Wiley Online Library, 2020.
  27. Accelerating 3d deep learning with pytorch3d. arXiv:2007.08501, 2020.
  28. Graph-based denoising for time-varying point clouds. In 2015 3DTV-Conference: The True Vision-Capture, Transmission and Display of 3D Video (3DTV-CON), pages 1–4. IEEE, 2015.
  29. Paris-rue-madame database: a 3d mobile laser scanner dataset for benchmarking urban detection, segmentation and classification methods. 2014.
  30. Generative modeling by estimating gradients of the data distribution. In Advances in Neural Information Processing Systems, pages 11918–11930, 2019.
  31. Denoising point sets via l0 minimization. Computer Aided Geometric Design, 35:2–15, 2015.
  32. Dynamic graph cnn for learning on point clouds. ACM Transactions on Graphics (TOG), 38(5):1–12, 2019.
  33. Survey on sparsity in geometric modeling and processing. Graphical Models, 82:160–180, 2015.
  34. Patch-based progressive 3d point set upsampling. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 5958–5967, 2019.
  35. Pu-net: Point cloud upsampling network. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 2790–2799, 2018.
  36. Density-based denoising of point cloud. In 9th International Conference on Robotic, Vision, Signal Processing and Power Applications, pages 287–295. Springer, 2017.
  37. 3D point cloud denoising using graph Laplacian regularization of a low dimensional manifold model. IEEE Transactions on Image Processing, 29:3474–3489, December 2019.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (2)
  1. Shitong Luo (17 papers)
  2. Wei Hu (309 papers)
Citations (134)
View on Semantic Scholar

Summary

Essay on "Score-Based Point Cloud Denoising"

The paper "Score-Based Point Cloud Denoising" by Shitong Luo and Wei Hu presents a novel approach to address the issue of noise inherent in point clouds produced by scanning devices. Noise in point clouds impedes subsequent tasks such as surface reconstruction and analysis, necessitating effective denoising methods. The paper introduces a score-based methodology for point cloud denoising, diverging significantly from traditional and contemporary techniques by leveraging probabilistic modeling and gradient ascent.

The authors posit that the distribution of noisy point clouds can be seen as the convolution of a noise-free sample distribution with a noise model, where the mode of this convolved distribution corresponds to the underlying clean surface. To perform denoising, the method increases the log-likelihood of each point from the convolved distribution using gradient ascent—a process that iteratively updates each point's position. Given that the convolved distribution is unknown at test-time, the approach involves a neural network model trained to estimate the score, i.e., the gradient of the log-probability function of the convolved distribution, based solely on the noisy point cloud input.

The paper introduces a detailed neural network architecture designed to learn and predict these scores, highlighting the focus on estimating point-wise scores on a local basis, which enhances generalizability across various shapes. The training procedure involves matching the neural network's output to a defined ground-truth score, formulated as the vector displacement towards the nearest clean surface point. This objective, which emphasizes the neighborhood of each point, is a distinct departure from previous approaches that focus solely on pointwise displacements.

Numerical results underscore the superiority of this method compared to existing state-of-the-art solutions. The proposed model was shown to outperform leading optimization-based methods and deep-learning-based denoisers across diverse noise settings. Crucially, despite being trained predominantly on Gaussian noise, the model exhibited robust generalizability by effectively handling simulated LiDAR noise, indicating its potential applicability across real-world scanning scenarios.

Implications of this research extend beyond denoising. The paper briefly explores point cloud upsampling by utilizing the denoising process on artificially noised data, demonstrating comparable or superior performance against specialized upsampling networks. This opens avenues for potential explorations into other point cloud processing tasks where mastery over detailed local geometry is requisite.

Future prospects could involve extending this work to incorporate dynamic or moving point clouds, such as those encountered in real-time 3D capture scenarios. Additionally, there may be potential in further developing unsupervised or semi-supervised adaptations to overcome the dependence on labeled data while still achieving effective denoising.

In summary, the paper presents a cogent framework for point cloud denoising predicated on distributional modeling and score-based optimization. Its demonstrated effectiveness, combined with its adaptability to other related applications, marks it as a significant contribution to the field of 3D data processing, offering both immediate applications and numerous directions for further research.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown
Github Logo Streamline Icon: https://streamlinehq.com

GitHub

  1. GitHub - luost26/score-denoise: :snowflake: Score-Based Point Cloud Denoising (ICCV 2021) (169 stars)