"Single Depth-image 3D Reflection Symmetry and Shape Prediction"
IEEE/CVF International Conference on Computer Vision (ICCV), 2023
Abstract
In this paper, we present Iterative Symmetry Completion Network (ISCNet), a single depth-image shape completion method that exploits reflective symmetry cues to obtain more detailed shapes. The efficacy of single depth-image shape completion methods is often sensitive to the accuracy of the symmetry plane. ISCNet therefore jointly estimates the symmetry plane and shape completion iteratively; more complete shapes contribute to more robust symmetry plane estimates and vice versa. Furthermore, our shape completion method operates in the image domain, enabling more efficient high-resolution, detailed geometry reconstruction. We perform the shape completion from pairs of viewpoints, reflected across the symmetry plane, predicted by a reinforcement learning agent to improve robustness and to simultaneously explicitly leverage symmetry. We demonstrate the effectiveness of ISCNet on a variety of object categories on both synthetic and real-scanned datasets.
Download
Supplementary Material
Bibtex
IEEE/CVF International Conference on Computer Vision (ICCV), 2023
AbstractIn this paper, we present Iterative Symmetry Completion Network (ISCNet), a single depth-image shape completion method that exploits reflective symmetry cues to obtain more detailed shapes. The efficacy of single depth-image shape completion methods is often sensitive to the accuracy of the symmetry plane. ISCNet therefore jointly estimates the symmetry plane and shape completion iteratively; more complete shapes contribute to more robust symmetry plane estimates and vice versa. Furthermore, our shape completion method operates in the image domain, enabling more efficient high-resolution, detailed geometry reconstruction. We perform the shape completion from pairs of viewpoints, reflected across the symmetry plane, predicted by a reinforcement learning agent to improve robustness and to simultaneously explicitly leverage symmetry. We demonstrate the effectiveness of ISCNet on a variety of object categories on both synthetic and real-scanned datasets.
Download
- Zhang_ICCV2023.pdf (3.52Mb)
Supplementary Material
- Document: Zhang_ICCV2023_supplemental.pdf (8.63Mb)
Bibtex
@conference{Zhang:2023:SDI,
author = {Zhang, Zhaoxuan and Dong, Bo and Li, Tong and Heide, Felix and Peers, Pieter and Yin, Baocai and Yang, Xin},
title = {Single Depth-image 3D Reflection Symmetry and Shape Prediction},
year = {2023},
booktitle = {IEEE/CVF International Conference on Computer Vision (ICCV)},
}
author = {Zhang, Zhaoxuan and Dong, Bo and Li, Tong and Heide, Felix and Peers, Pieter and Yin, Baocai and Yang, Xin},
title = {Single Depth-image 3D Reflection Symmetry and Shape Prediction},
year = {2023},
booktitle = {IEEE/CVF International Conference on Computer Vision (ICCV)},
}