Japanese / English

Detail of Publication

Text Language English
Authors Fairuz Safwan Mahad, Masakazu Iwamura and Koichi Kise
Title Leveraging Pyramidal Feature Hierarchy for 3D Reconstruction
Journal Proc. International Workshop on Frontiers of Computer Vision (IW-FCV)
Reviewed or not Reviewed
Presentation type Oral
Month & Year February 2020
Abstract 3D reconstruction methods with neural networks has been on a rising trend and extensive research has been done. However, they do not pay particular attention to reconstructing the detailed parts of the objects. Although the reconstructed 3D models often look complete generally, close-up observations reveal that most of them lose out on the detailed parts, which reduces the quality of the 3D model. We observe that especially fine parts of the object either fail to be reconstructed or are less dense and incomplete. This happens because the network is not designed to capture the fine details of the object. Therefore, in this paper, we propose a network which is designed to capture both the coarse and fine details of the object in an effort to improve the reconstruction of the fine parts of the object. In order to do so, we design a multi-branch deep generative network which learns the local features, generic features and also intermediate features. Utilizing these features allows the network to learn features at different levels which are able to reconstruct the fine parts as well as the overall shape of the reconstructed 3D model. We show that our method outperformed the state-of-the-art.
DOI 10.1007/978-981-15-4818-5_26
Back to list