Points to patches: Enabling the use of self-attention for 3D shape recognition
(2022) 26TH International Conference on Pattern Recognition, 2022 In International Conference on Pattern Recognition p.528-534- Abstract
- While the Transformer architecture has become ubiquitous in the machine learning field, its adaptation to 3D shape recognition is non-trivial. Due to its quadratic computational complexity, the self-attention operator quickly becomes inefficient as the set of input points grows larger. Furthermore, we find that the attention mechanism struggles to find useful connections between individual points on a global scale. In order to alleviate these problems, we propose a two-stage Point Transformer-in-Transformer (Point-TnT) approach which combines local and global attention mechanisms, enabling both individual points and patches of points to attend to each other effectively. Experiments on shape classification show that such an approach... (More)
- While the Transformer architecture has become ubiquitous in the machine learning field, its adaptation to 3D shape recognition is non-trivial. Due to its quadratic computational complexity, the self-attention operator quickly becomes inefficient as the set of input points grows larger. Furthermore, we find that the attention mechanism struggles to find useful connections between individual points on a global scale. In order to alleviate these problems, we propose a two-stage Point Transformer-in-Transformer (Point-TnT) approach which combines local and global attention mechanisms, enabling both individual points and patches of points to attend to each other effectively. Experiments on shape classification show that such an approach provides more useful features for downstream tasks than the baseline Transformer, while also being more computationally efficient. In addition, we also extend our method to feature matching for scene reconstruction, showing that it can be used in conjunction with existing scene reconstruction pipelines. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/981ccdb2-f80d-4ce4-b744-690fe33f6107
- author
- Berg, Axel LU ; Oskarsson, Magnus LU and O'Connor, Mark
- organization
- publishing date
- 2022-08-21
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- host publication
- 2022 26th International Conference on Pattern Recognition (ICPR)
- series title
- International Conference on Pattern Recognition
- pages
- 7 pages
- publisher
- IEEE - Institute of Electrical and Electronics Engineers Inc.
- conference name
- 26TH International Conference on Pattern Recognition, 2022
- conference location
- Montreal, Canada
- conference dates
- 2022-08-21 - 2022-08-25
- external identifiers
-
- scopus:85143583530
- ISSN
- 1051-4651
- 2831-7475
- ISBN
- 978-1-6654-9062-7
- 978-1-6654-9062-7
- DOI
- 10.48550/arXiv.2204.03957
- project
- Deep Learning for Simultaneous Localization and Mapping
- language
- English
- LU publication?
- yes
- id
- 981ccdb2-f80d-4ce4-b744-690fe33f6107
- date added to LUP
- 2022-12-16 08:32:51
- date last changed
- 2024-06-29 01:36:29
@inproceedings{981ccdb2-f80d-4ce4-b744-690fe33f6107, abstract = {{While the Transformer architecture has become ubiquitous in the machine learning field, its adaptation to 3D shape recognition is non-trivial. Due to its quadratic computational complexity, the self-attention operator quickly becomes inefficient as the set of input points grows larger. Furthermore, we find that the attention mechanism struggles to find useful connections between individual points on a global scale. In order to alleviate these problems, we propose a two-stage Point Transformer-in-Transformer (Point-TnT) approach which combines local and global attention mechanisms, enabling both individual points and patches of points to attend to each other effectively. Experiments on shape classification show that such an approach provides more useful features for downstream tasks than the baseline Transformer, while also being more computationally efficient. In addition, we also extend our method to feature matching for scene reconstruction, showing that it can be used in conjunction with existing scene reconstruction pipelines.}}, author = {{Berg, Axel and Oskarsson, Magnus and O'Connor, Mark}}, booktitle = {{2022 26th International Conference on Pattern Recognition (ICPR)}}, isbn = {{978-1-6654-9062-7}}, issn = {{1051-4651}}, language = {{eng}}, month = {{08}}, pages = {{528--534}}, publisher = {{IEEE - Institute of Electrical and Electronics Engineers Inc.}}, series = {{International Conference on Pattern Recognition}}, title = {{Points to patches: Enabling the use of self-attention for 3D shape recognition}}, url = {{http://dx.doi.org/10.48550/arXiv.2204.03957}}, doi = {{10.48550/arXiv.2204.03957}}, year = {{2022}}, }