3D Mesh Segmentation via Multi-branch 1D Convolutional Neural Networks
- Submitting institution
-
Swansea University / Prifysgol Abertawe
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 38364
- Type
- D - Journal article
- DOI
-
10.1016/j.gmod.2018.01.001
- Title of journal
- Graphical Models
- Article number
- -
- First page
- 1
- Volume
- 96
- Issue
- -
- ISSN
- 15240703
- Open access status
- Compliant
- Month of publication
- March
- Year of publication
- 2018
- URL
-
-
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
-
- Research group(s)
-
-
- Citation count
- 7
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- 3D segmentation is an important but hard problem for shape understanding due to its semantic/functional nature. There were few machine/deep-learning techniques, and most previous work provided no source codes affecting subsequent development/comparison. This study innovated a robust conformal factor feature, and a feature-based multi-branch 1D CNN technique for 3D mesh segmentation that beats previous techniques. The study offers a comprehensive evaluation against 3 machine/deep-learning architectures. All 4 implementations and data are made publicly available. We further raise the inaccurate labelling problem of existing public ground truth. These insights further inspired a new active learning annotation technique using fast feature-based learning.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -