File Download
There are no files associated with this item.
Links for fulltext
(May Require Subscription)
- Publisher Website: 10.1109/TMM.2022.3198011
- Scopus: eid_2-s2.0-85135970280
- WOS: WOS:001098831500001
- Find via
Supplementary
- Citations:
- Appears in Collections:
Article: GeometryMotion-Transformer: An End-to-End Framework for 3D Action Recognition
Title | GeometryMotion-Transformer: An End-to-End Framework for 3D Action Recognition |
---|---|
Authors | |
Keywords | 3D action recognition Feature extraction Finite element analysis Geometry point cloud Point cloud compression Task analysis Three-dimensional displays transformer Transformers |
Issue Date | 2022 |
Citation | IEEE Transactions on Multimedia, 2022 How to Cite? |
Abstract | In this work, we propose a new end-to-end optimized two-stream framework called GeometryMotion-Transformer (GMT) for 3D action recognition. We first observe that the existing 3D action recognition approaches cannot well extract motion representations from point cloud sequences. Specifically, when extracting motion representations, the existing approaches do not explicitly consider one-to-one correspondence among frames. Besides, the existing methods only extract the |
Persistent Identifier | http://hdl.handle.net/10722/322003 |
ISSN | 2023 Impact Factor: 8.4 2023 SCImago Journal Rankings: 2.260 |
ISI Accession Number ID |
DC Field | Value | Language |
---|---|---|
dc.contributor.author | Liu, Jiaheng | - |
dc.contributor.author | Guo, Jinyang | - |
dc.contributor.author | Xu, Dong | - |
dc.date.accessioned | 2022-11-03T02:22:56Z | - |
dc.date.available | 2022-11-03T02:22:56Z | - |
dc.date.issued | 2022 | - |
dc.identifier.citation | IEEE Transactions on Multimedia, 2022 | - |
dc.identifier.issn | 1520-9210 | - |
dc.identifier.uri | http://hdl.handle.net/10722/322003 | - |
dc.description.abstract | In this work, we propose a new end-to-end optimized two-stream framework called GeometryMotion-Transformer (GMT) for 3D action recognition. We first observe that the existing 3D action recognition approaches cannot well extract motion representations from point cloud sequences. Specifically, when extracting motion representations, the existing approaches do not explicitly consider one-to-one correspondence among frames. Besides, the existing methods only extract the <italic>single-scale</italic> motion representations, which cannot well model the complex motion patterns of moving objects in point cloud sequences. To address these issues, we first propose the feature extraction module (FEM) to generate one-to-one correspondence among frames without using the voxelization process, and explicitly extract both geometry and <italic>multi-scale</italic> motion representations from raw point clouds. Moreover, we also observe the existing two-stream 3D action recognition approaches simply concatenate or add the geometry and motion features, which cannot well exploit the relationship between two-steam features. To this end, we also propose an improved transformer-based feature fusion module (FFM) to effectively fuse the two-stream features. Based on the proposed FEM and FFM, we build our GMT for 3D action recognition. Extensive experimental results on four benchmark datasets demonstrate the effectiveness of our backbone GMT. | - |
dc.language | eng | - |
dc.relation.ispartof | IEEE Transactions on Multimedia | - |
dc.subject | 3D action recognition | - |
dc.subject | Feature extraction | - |
dc.subject | Finite element analysis | - |
dc.subject | Geometry | - |
dc.subject | point cloud | - |
dc.subject | Point cloud compression | - |
dc.subject | Task analysis | - |
dc.subject | Three-dimensional displays | - |
dc.subject | transformer | - |
dc.subject | Transformers | - |
dc.title | GeometryMotion-Transformer: An End-to-End Framework for 3D Action Recognition | - |
dc.type | Article | - |
dc.description.nature | link_to_subscribed_fulltext | - |
dc.identifier.doi | 10.1109/TMM.2022.3198011 | - |
dc.identifier.scopus | eid_2-s2.0-85135970280 | - |
dc.identifier.eissn | 1941-0077 | - |
dc.identifier.isi | WOS:001098831500001 | - |