Graph Neural Networks for Point Cloud Processing

3D Point clouds are a rich source of information that enjoy growing popularity in the vision community. However processing unordered and sparse point clouds using neural networks has been a challenge but in recent years there are models proposed to learn point clouds using deep learning. Graph neural networks have also shown great capacity to capture geometrical features from point clouds in tasks such as classification and segmentation. In this presentation we discuss how graphs can be utilized to describe point cloud patches, detect salient points and use them in downstream tasks such as 3D registration.
Lecture slides: drive.google.com/file/d/12Jgf...
00:00 Intro
03:02 Depth map
04:19 Voxels
07:38 Point Clouds
12:46 Graphs
13:41 EdgeConv
16:00 Point Cloud Registration - Graphite: GRAPH-Induced feaTure Extraction for Point Cloud Registration (3DV 2020)
17:38 Global methods
19:29 Local methods - Point Clouds, Voxels
21:15 Graph construction
21:51 Architecture
27:42 Training with pose variations
28:16 ModelNet40 registration
30:09 3dmatch registration
30:42 Geometric registration benchmark
31:24 Q/A
40:39 Graph-based Matching - CoFiNet: Reliable Coarse-to-fine Correspondences for Robust Point Cloud (NeurIPS 2021)
42:15 Sampling and description
43:41 Context Aggregation using GNN
47:43 Proposal Refinement Stage
48:41 Results
49:37 Conclusion
50:28 Q/A and discussion
[Chapters were auto-generated using our proprietary software - contact us if you are interested in access to the software]
Talk is based on the speaker's paper:
Graphite: GRAPH-Induced feaTure Extraction for Point Cloud Registration
arxiv.org/abs/2010.09079
Presenter BIO:
Mahdi Saleh studied Bachelor of Electrical Engineering at IUST. He then moved to Germany to study for his Master's at the Technical University of Munich Computer Science department. Meanwhile, he was working as a researcher in 3D computer vision and AI in Framos GmbH, IBM Watson Munich, and AR Experts. Before starting his Ph.D. at TUM, he worked on industrial computer vision and applied research for two years. He is now a Ph.D. student at the CV group of the CAMP chair at TUM focused on Point cloud processing and 3D pose estimation. At the moment, he is also a research Scientist intern at Facebook Reality Lab, Menlo Park.
-------------------------
Find us at:
Newsletter for updates about more events ➜ eepurl.com/gJ1t-D
Sub-reddit for discussions ➜ / 2d3dai
Discord server for, well, discord ➜ / discord
Blog ➜ 2d3d.ai
We are recruiting software and ML engineers - message me for further details

[@2d3dai]

Would love to hear any comment or input you have. Did you find the content interesting? Is there a place for improvement? Please leave any feedback you have here in a comment :)