Dynamic Graph CNN for Learning on Point Clouds
Point clouds provide a flexible geometric representation suitable for countless applications in computer graphics; they also comprise the raw output of most 3D data acquisition devices. While hand-designed features on point clouds have long been proposed in graphics and vision, however, the recent o...
| Main Authors: | , , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
Association for Computing Machinery (ACM)
2020
|
| Online Access: | https://hdl.handle.net/1721.1/126819 |
| _version_ | 1826205336374083584 |
|---|---|
| author | Wang, Yue Sun, Yongbin Liu, Ziwei Sarma, Sanjay E |
| author2 | Massachusetts Institute of Technology. Department of Mechanical Engineering |
| author_facet | Massachusetts Institute of Technology. Department of Mechanical Engineering Wang, Yue Sun, Yongbin Liu, Ziwei Sarma, Sanjay E |
| author_sort | Wang, Yue |
| collection | MIT |
| description | Point clouds provide a flexible geometric representation suitable for countless applications in computer graphics; they also comprise the raw output of most 3D data acquisition devices. While hand-designed features on point clouds have long been proposed in graphics and vision, however, the recent overwhelming success of convolutional neural networks (CNNs) for image analysis suggests the value of adapting insight from CNN to the point cloud world. Point clouds inherently lack topological information, so designing a model to recover topology can enrich the representation power of point clouds. To this end, we propose a new neural network module dubbed EdgeConv suitable for CNN-based high-level tasks on point clouds, including classification and segmentation. EdgeConv acts on graphs dynamically computed in each layer of the network. It is differentiable and can be plugged into existing architectures. Compared to existing modules operating in extrinsic space or treating each point independently, EdgeConv has several appealing properties: It incorporates local neighborhood information; it can be stacked applied to learn global shape properties; and in multi-layer systems affinity in feature space captures semantic characteristics over potentially long distances in the original embedding. We show the performance of our model on standard benchmarks, including ModelNet40, ShapeNetPart, and S3DIS. |
| first_indexed | 2024-09-23T13:11:05Z |
| format | Article |
| id | mit-1721.1/126819 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T13:11:05Z |
| publishDate | 2020 |
| publisher | Association for Computing Machinery (ACM) |
| record_format | dspace |
| spelling | mit-1721.1/1268192022-10-01T13:38:24Z Dynamic Graph CNN for Learning on Point Clouds Wang, Yue Sun, Yongbin Liu, Ziwei Sarma, Sanjay E Massachusetts Institute of Technology. Department of Mechanical Engineering Massachusetts Institute of Technology. Materials Research Laboratory Point clouds provide a flexible geometric representation suitable for countless applications in computer graphics; they also comprise the raw output of most 3D data acquisition devices. While hand-designed features on point clouds have long been proposed in graphics and vision, however, the recent overwhelming success of convolutional neural networks (CNNs) for image analysis suggests the value of adapting insight from CNN to the point cloud world. Point clouds inherently lack topological information, so designing a model to recover topology can enrich the representation power of point clouds. To this end, we propose a new neural network module dubbed EdgeConv suitable for CNN-based high-level tasks on point clouds, including classification and segmentation. EdgeConv acts on graphs dynamically computed in each layer of the network. It is differentiable and can be plugged into existing architectures. Compared to existing modules operating in extrinsic space or treating each point independently, EdgeConv has several appealing properties: It incorporates local neighborhood information; it can be stacked applied to learn global shape properties; and in multi-layer systems affinity in feature space captures semantic characteristics over potentially long distances in the original embedding. We show the performance of our model on standard benchmarks, including ModelNet40, ShapeNetPart, and S3DIS. Army Research Office (Grant W911NF-12-R-0011) Air Force Office of Scientific Research (Award FA9550-19-1-0319) National Science Foundation (Grant IIS-1838071) 2020-08-26T15:31:08Z 2020-08-26T15:31:08Z 2019-10 2020-08-04T18:37:47Z Article http://purl.org/eprint/type/JournalArticle https://hdl.handle.net/1721.1/126819 Wang, Yue et al. "Dynamic Graph CNN for Learning on Point Clouds." ACM Transactions on Graphics 38, 5 (October 2019): 146 © 2019 The Author(s) en http://dx.doi.org/10.1145/3326362 ACM Transactions on Graphics Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf Association for Computing Machinery (ACM) arXiv |
| spellingShingle | Wang, Yue Sun, Yongbin Liu, Ziwei Sarma, Sanjay E Dynamic Graph CNN for Learning on Point Clouds |
| title | Dynamic Graph CNN for Learning on Point Clouds |
| title_full | Dynamic Graph CNN for Learning on Point Clouds |
| title_fullStr | Dynamic Graph CNN for Learning on Point Clouds |
| title_full_unstemmed | Dynamic Graph CNN for Learning on Point Clouds |
| title_short | Dynamic Graph CNN for Learning on Point Clouds |
| title_sort | dynamic graph cnn for learning on point clouds |
| url | https://hdl.handle.net/1721.1/126819 |
| work_keys_str_mv | AT wangyue dynamicgraphcnnforlearningonpointclouds AT sunyongbin dynamicgraphcnnforlearningonpointclouds AT liuziwei dynamicgraphcnnforlearningonpointclouds AT sarmasanjaye dynamicgraphcnnforlearningonpointclouds |