A Shared Pose Regression Network for Pose Estimation of Objects from RGB Images
(2022) IEEE/RSJ International Conference on Signal Image Technology & Internet Based Systems (SITIS) 2022- Abstract
- In this paper we propose a shared regression network to jointly estimate the pose of multiple objects, replacing multiple object-specific solutions. We demonstrate that this shared network can outperform other similar approaches that rely on multiple object-specific models by evaluating it on the TLESS dataset using the VSD (Visible Surface Discrepancy). Our approach offers a less complex solution, with fewer parameters, lower memory consumption and less training required. Furthermore, it inherently handles symmetric objects by using a depth-based loss during training and can predict in real-time. Finally, we show how our proposed pipeline can be used for fine-tuning a feature extractor jointly on all objects while training the shared pose... (More)
- In this paper we propose a shared regression network to jointly estimate the pose of multiple objects, replacing multiple object-specific solutions. We demonstrate that this shared network can outperform other similar approaches that rely on multiple object-specific models by evaluating it on the TLESS dataset using the VSD (Visible Surface Discrepancy). Our approach offers a less complex solution, with fewer parameters, lower memory consumption and less training required. Furthermore, it inherently handles symmetric objects by using a depth-based loss during training and can predict in real-time. Finally, we show how our proposed pipeline can be used for fine-tuning a feature extractor jointly on all objects while training the shared pose regression network. This fine-tuning process improves the pose estimation performance. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/1c75c100-5dc0-465a-85c3-d076f7b584a1
- author
- Hein Bengtson, Stefan LU ; Åström, Hampus LU ; Moeslund, Thomas B. ; Topp, Elin A. LU and Krueger, Volker LU
- organization
- publishing date
- 2022-10
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- keywords
- Training, Measurement, Solid modeling, image resolution, Pose estimation, neural networks, memory management
- host publication
- IEEE/RSJ International Conference on Signal Image Technology & Internet Based Systems (SITIS)
- pages
- 8 pages
- publisher
- IEEE - Institute of Electrical and Electronics Engineers Inc.
- conference name
- IEEE/RSJ International Conference on Signal Image Technology & Internet Based Systems (SITIS) 2022
- conference location
- Dijon, France
- conference dates
- 2022-10-19 - 2022-10-21
- external identifiers
-
- scopus:85154031291
- ISBN
- 978-1-6654-6496-3
- 978-1-6654-6495-6
- DOI
- 10.1109/SITIS57111.2022.00022
- project
- RobotLab LTH
- language
- English
- LU publication?
- yes
- id
- 1c75c100-5dc0-465a-85c3-d076f7b584a1
- date added to LUP
- 2023-07-20 16:38:08
- date last changed
- 2024-06-01 04:15:52
@inproceedings{1c75c100-5dc0-465a-85c3-d076f7b584a1, abstract = {{In this paper we propose a shared regression network to jointly estimate the pose of multiple objects, replacing multiple object-specific solutions. We demonstrate that this shared network can outperform other similar approaches that rely on multiple object-specific models by evaluating it on the TLESS dataset using the VSD (Visible Surface Discrepancy). Our approach offers a less complex solution, with fewer parameters, lower memory consumption and less training required. Furthermore, it inherently handles symmetric objects by using a depth-based loss during training and can predict in real-time. Finally, we show how our proposed pipeline can be used for fine-tuning a feature extractor jointly on all objects while training the shared pose regression network. This fine-tuning process improves the pose estimation performance.}}, author = {{Hein Bengtson, Stefan and Åström, Hampus and Moeslund, Thomas B. and Topp, Elin A. and Krueger, Volker}}, booktitle = {{IEEE/RSJ International Conference on Signal Image Technology & Internet Based Systems (SITIS)}}, isbn = {{978-1-6654-6496-3}}, keywords = {{Training; Measurement; Solid modeling; image resolution; Pose estimation; neural networks; memory management}}, language = {{eng}}, publisher = {{IEEE - Institute of Electrical and Electronics Engineers Inc.}}, title = {{A Shared Pose Regression Network for Pose Estimation of Objects from RGB Images}}, url = {{http://dx.doi.org/10.1109/SITIS57111.2022.00022}}, doi = {{10.1109/SITIS57111.2022.00022}}, year = {{2022}}, }