Scholarlink

<p>In this paper, we introduce a new framework for single view 3D mesh reconstruction. It allows to generate annotated 2D synthetic data from readily available 3D models. The generated data are realistic and diverse, to tackle the domain shift problem. The efficiency of the database is demonstrated using synthetic data generated to train two common mesh reconstruction networks. Additionally, we incorporate additional metadata from our database to provide weak 3D supervision loss function during training. The proposed loss function can be easily embedded in training process of mesh reconstruction networks. We evaluated our proposed framework on a dataset of real images and compared our results with those of previous works.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

3D mesh reconstruction of objects from a single view using machine vision and synthetic data

  • 期刊论文

  • 发表日期：

    2025年12月8日

查看全文
  • Sohrab Moayer,
  • Maryam Shoaran

摘要

In this paper, we introduce a new framework for single view 3D mesh reconstruction. It allows to generate annotated 2D synthetic data from readily available 3D models. The generated data are realistic and diverse, to tackle the domain shift problem. The efficiency of the database is demonstrated using synthetic data generated to train two common mesh reconstruction networks. Additionally, we incorporate additional metadata from our database to provide weak 3D supervision loss function during training. The proposed loss function can be easily embedded in training process of mesh reconstruction networks. We evaluated our proposed framework on a dataset of real images and compared our results with those of previous works.