RealImage2Mesh: Reconstructing symmetrical mesh models from single real-world images

Ying Li, Yue Yu*, Bowen Li, Yue Yang

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

We propose a method to reconstruct mesh models from single real-world images. Our method takes a real-world image of an object as input, uses an encoder-decoder network to extract 2. 5D information such as the silhouette, depth, and normals of the object, and then inputs this information into a multi-level 3D mesh generation network to generate a high-quality mesh model using mesh deformation and face pruning. In the process of generating the mesh model, we propose a symmetry loss to keep the shape of the mesh model reasonable. According to our experiments, compared with other single-view object reconstruction methods, our method can reconstruct 3D models with higher quality from real-world images.

Original languageEnglish
Title of host publicationICCSE 2021 - IEEE 16th International Conference on Computer Science and Education
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages333-338
Number of pages6
ISBN (Electronic)9781665414685
DOIs
Publication statusPublished - 17 Aug 2021
Event16th IEEE International Conference on Computer Science and Education, ICCSE 2021 - Lancaster, United Kingdom
Duration: 17 Aug 202121 Aug 2021

Publication series

NameICCSE 2021 - IEEE 16th International Conference on Computer Science and Education

Conference

Conference16th IEEE International Conference on Computer Science and Education, ICCSE 2021
Country/TerritoryUnited Kingdom
CityLancaster
Period17/08/2121/08/21

Keywords

  • Mesh model
  • Real-world image
  • Single-view reconstruction

Fingerprint

Dive into the research topics of 'RealImage2Mesh: Reconstructing symmetrical mesh models from single real-world images'. Together they form a unique fingerprint.

Cite this