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

PosDiffNet: Positional Neural Diffusion for Point Cloud Registration in a Large Field of View with Perturbations (2401.03167v1)

Published 6 Jan 2024 in cs.CV and cs.AI

Abstract: Point cloud registration is a crucial technique in 3D computer vision with a wide range of applications. However, this task can be challenging, particularly in large fields of view with dynamic objects, environmental noise, or other perturbations. To address this challenge, we propose a model called PosDiffNet. Our approach performs hierarchical registration based on window-level, patch-level, and point-level correspondence. We leverage a graph neural partial differential equation (PDE) based on Beltrami flow to obtain high-dimensional features and position embeddings for point clouds. We incorporate position embeddings into a Transformer module based on a neural ordinary differential equation (ODE) to efficiently represent patches within points. We employ the multi-level correspondence derived from the high feature similarity scores to facilitate alignment between point clouds. Subsequently, we use registration methods such as SVD-based algorithms to predict the transformation using corresponding point pairs. We evaluate PosDiffNet on several 3D point cloud datasets, verifying that it achieves state-of-the-art (SOTA) performance for point cloud registration in large fields of view with perturbations. The implementation code of experiments is available at https://github.com/AI-IT-AVs/PosDiffNet.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (72)
  1. BUFFER: Balancing Accuracy, Efficiency, and Generalizability in Point Cloud Registration. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 1255–1264.
  2. SpinNet: Learning a general surface descriptor for 3d point cloud registration. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 11753–11762.
  3. Pointdsc: Robust point cloud registration using deep spatial consistency. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 15859–15869.
  4. D3feat: Joint learning of dense detection and description of 3d local features. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 6359–6367.
  5. Method for registration of 3-D shapes. In Sensor fusion IV: control paradigms and data structures, volume 1611, 586–606.
  6. Boreas: A multi-season autonomous driving dataset. Int. J. Robot. Res., 42(1-2): 33–42.
  7. Beltrami flow and neural diffusion on graphs. Adv. Neural Inform. Process. Syst., 34: 1594–1609.
  8. GRAND: Graph Neural Diffusion. In Proc. Int. Conf. Mach. Learn.
  9. Neural ordinary differential equations. Adv. Neural Inform. Process. Syst., 31.
  10. Chen, R. T. Q. 2018. torchdiffeq.
  11. SC2-PCR: A second order spatial compatibility for efficient and robust point cloud registration. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 13221–13231.
  12. Fully convolutional geometric features. In Proc. Int. Conf. Comput. Vis., 8958–8966.
  13. Augmented neural odes. Adv. Neural Inform. Process. Syst., 32.
  14. Hypergraph neural networks. In Proc. AAAI Conf. Artificial Intell., 3558–3565.
  15. Stickypillars: Robust and efficient feature matching on point clouds using graph neural networks. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 313–323.
  16. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography. Communications of the ACM, 24(6): 381–395.
  17. Vision meets robotics: The KITTI dataset. Int. J. Rob. Res., 32(11): 1231–1237.
  18. Regularisation of neural networks by enforcing lipschitz continuity. Mach. Learn., 110: 393–416.
  19. Mixed integer linear programming for quality of service optimization in Clouds. Future Generation Computer Systems, 71: 1–17.
  20. PCT: Point cloud transformer. Comput. Vis. Media., 7: 187–199.
  21. Predator: Registration of 3d point clouds with low overlap. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 4267–4276.
  22. Location learning for AVs: LiDAR and image landmarks fusion localization with graph neural networks. In Proc. IEEE Int. Conf. Intell. Transp. Syst., 3032–3037.
  23. Stable neural ODE with Lyapunov-stable equilibrium points for defending against adversarial attacks. Adv. Neural Inform. Process. Syst.
  24. Introduction to the modern theory of dynamical systems. 54. Cambridge university press.
  25. From high energy physics to low level vision. In Proc. Int. Conf. Scale-Space Theory Comput. Vision, 236–247.
  26. Adam: A method for stochastic optimization. In Proc. Int. Conf. Learn. Represent., 1–15.
  27. Voxelized GICP for fast and accurate 3d point cloud registration. In Proc. IEEE Int. Conf. Robot. Autom., 11054–11059.
  28. Resource efficient 3d convolutional neural networks. In Proc. Int. Conf. Comput. Vis. Worksh., 0–0.
  29. LP-3DCNN: Unveiling local phase in 3d convolutional neural networks. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 4903–4912.
  30. Accelerating neural odes using model order reduction. IEEE Trans. Neural Netw. Learn. Syst.
  31. So-net: Self-organizing network for point cloud analysis. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 9397–9406.
  32. OpenStreetMap-based autonomous navigation for the four wheel-legged robot via 3D-LiDAR and CCD camera. IEEE Trans. Ind. Electron., 69(3): 2708–2717.
  33. Lepard: Learning partial point cloud matching in rigid and deformable scenes. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 5554–5564.
  34. HRegNet: A hierarchical network for large-scale outdoor LiDAR point cloud registration. In Proc. Int. Conf. Comput. Vis., 16014–16023.
  35. Rethinking network design and local geometry in point cloud: A simple residual MLP framework. In Proc. Int. Conf. Learn. Represent.
  36. Unsupervised Deep Probabilistic Approach for Partial Point Cloud Registration. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 13611–13620.
  37. 3dregnet: A deep neural network for 3d point registration. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 7193–7203.
  38. Opening the blackbox: Accelerating neural differential equations by regularizing internal solver heuristics. In Proc. Int. Conf. Mach. Learn., 8325–8335. PMLR.
  39. Learning general and distinctive 3D local deep descriptors for point cloud registration. IEEE Trans. Pattern Anal. Mach. Intell.
  40. PointNet: Deep learning on point sets for 3d classification and segmentation. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 652–660.
  41. PointNeXt: Revisiting pointnet++ with improved training and scaling strategies. Adv. Neural Inform. Process. Syst., 35: 23192–23204.
  42. Geometric transformer for fast and robust point cloud registration. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 11143–11152.
  43. Compatibility-guided sampling consensus for 3-d point cloud registration. IEEE Trans. Geosci. Remote. Sens., 58(10): 7380–7392.
  44. Fast point feature histograms (FPFH) for 3D registration. In Proc. Proc. IEEE Int. Conf. Robot. Automat., 3212–3217.
  45. Superglue: Learning feature matching with graph neural networks. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 4938–4947.
  46. Generalized-ICP. In Robotics: science and systems, 435. Seattle, WA.
  47. LIO-SAM: Tightly-coupled LiDAR inertial odometry via smoothing and mapping. In Proc. IEEE Int. Conf. Intell. Robot. Syst., 5135–5142.
  48. RobustMat: Neural diffusion for street landmark patch matching under challenging environments. IEEE Trans. Image Process., 32: 5550–5563.
  49. Mvx-net: Multimodal voxelnet for 3d object detection. In Proc. IEEE Int. Conf. Robot. Autom., 7276–7282.
  50. Snow, D. R. 1972. Gronwall’s inequality for systems of partial differential equations in two independent variables. Proc. American Mathematical Soc., 33(1): 46–54.
  51. On the robustness of graph neural diffusion to topology perturbations. Adv. Neural Inform. Process. Syst., 35: 6384–6396.
  52. Multi-view convolutional neural networks for 3d shape recognition. In Proc. Int. Conf. Comput. Vis., 945–953.
  53. LoFTR: Detector-free local feature matching with transformers. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 8922–8931.
  54. Kpconv: Flexible and deformable convolution for point clouds. In Proc. Int. Conf. Comput. Vis., 6411–6420.
  55. Attention is all you need. Adv. Neural Inform. Process. Syst., 30.
  56. Graph attention networks. In Proc. Int. Conf. Learn. Represent., 1–12.
  57. Atloc: Attention guided camera localization. In Proc. AAAI Conf. Artificial Intell., 10393–10401.
  58. RoReg: Pairwise Point Cloud Registration with Oriented Descriptors and Local Rotations. IEEE Trans. Pattern Anal. Mach. Intell.
  59. RobustLoc: Robust Camera Pose Regression in Challenging Driving Environments. In Proc. AAAI Conf. Artificial Intell.
  60. HypLiLoc: Towards Effective LiDAR Pose Regression with Hyperbolic Fusion. In Proc. IEEE Conf. Comput. Vis. Pattern Recog.
  61. Deep closest point: Learning representations for point cloud registration. In Proc. Int. Conf. Comput. Vis., 3523–3532.
  62. Dynamic graph cnn for learning on point clouds. ACM Trans. Graphics, 38(5): 1–12.
  63. End-to-end 3d point cloud learning for registration task using virtual correspondences. In Proc. IEEE Int. Conf. Intell. Robot. Syst., 2678–2683.
  64. On robustness of neural ordinary differential equations. Adv. Neural Inform. Process. Syst., 1–13.
  65. CoFiNet: Reliable coarse-to-fine correspondences for robust pointcloud registration. Adv. Neural Inform. Process. Syst., 34: 23872–23884.
  66. Rotation-invariant transformer for point cloud matching. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 5384–5393.
  67. An advanced outlier detected total least-squares algorithm for 3-D point clouds registration. IEEE Trans. Geosci. Remote. Sens., 57(7): 4789–4798.
  68. Parameter is not all you need: Starting from non-parametric networks for 3d point cloud analysis. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 5344–5353.
  69. 3D Registration with Maximal Cliques. In Proc. IEEE Conf. Comput. Vis. Pattern Recog., 17745–17754.
  70. Learning a task-specific descriptor for robust matching of 3d point clouds. IEEE Trans. Circuits Syst. Video Technol., 32(12): 8462–8475.
  71. Adversarial robustness in graph neural networks: A Hamiltonian approach. Adv. Neural Inform. Process. Syst.
  72. SuperLine3D: Self-supervised Line Segmentation and Description for LiDAR Point Cloud. In Proc. Eur. Conf. Comput. Vis., 263–279.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (8)
  1. Rui She (37 papers)
  2. Sijie Wang (21 papers)
  3. Qiyu Kang (25 papers)
  4. Kai Zhao (160 papers)
  5. Yang Song (299 papers)
  6. Wee Peng Tay (101 papers)
  7. Tianyu Geng (2 papers)
  8. Xingchao Jian (15 papers)
Citations (2)
View on Semantic Scholar

Summary

We haven't generated a summary for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize Now
Github Logo Streamline Icon: https://streamlinehq.com

GitHub

  1. GitHub - AI-IT-AVs/PosDiffNet: The code for PosDiffNet (12 stars)
X Twitter Logo Streamline Icon: https://streamlinehq.com

Tweets