- 115 views
- 270 downloads
Animal 3D reconstruction from videos
-
- Author / Creator
- Ma, Youdong
-
3D reconstruction of quadruped animals is a challenging problem, where key issues
lie in their large shape variety and deformation within the same animal species as
well as the lack of sufficient training data. In this thesis, we present two approaches
toward this task. Our first approach is a model-based method called SMALF, where a
parametric 3D shape template, SMAL, is employed to recover the 3D pose and shape
of an animal from its raw video. On top of SMAL, our predicted mesh is further
allowed to perform per-vertex deformations: this way, our proposed model benefits
from the use of both parametric and non-parametric representations. Second, we
present AnimalRecon, yet another method that uses a neural implicit function to
represent 3D animal shapes. Neural implicit representation methods enjoy a higher
degree of flexibility and expressiveness due to the continuous nature of this shape
representation. However, how to extract a space-time coherent representation for
a video can be difficult to achieve. First, we implement neural blend skinning, a
method to enable our implicit shape to deform. Our implicit shape model is then
fitted into the given video. Qualitative and quantitative results of our approach
SMALF are conducted on two BADJA and RGBD-Dog datasets, where our method
is demonstrated to improve over the baselines. Our method AnimalRecon is further
examined on the RGBD-Dog dataset, where higher-fidelity 3D reconstruction results
are shown when comparing with the existing efforts. -
- Subjects / Keywords
-
- Graduation date
- Spring 2023
-
- Type of Item
- Thesis
-
- Degree
- Master of Science
-
- License
- This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.