What Do Single-view 3D Reconstruction Networks Learn?

被引:274
|
作者
Tatarchenko, Maxim [1 ]
Richter, Stephan R. [2 ]
Ranftl, Rene [2 ]
Li, Zhuwen [2 ]
Koltun, Vladlen [2 ]
Brox, Thomas [1 ]
机构
[1] Univ Freiburg, Freiburg, Germany
[2] Intel Labs, Santa Clara, CA USA
关键词
SHAPE;
D O I
10.1109/CVPR.2019.00352
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Convolutional networks for single-view object reconstruction have shown impressive performance and have become a popular subject of research. All existing techniques are united by the idea of having an encoder-decoder network that performs non-trivial reasoning about the 3D structure of the output space. In this work, we set up two alternative approaches that perform image classification and retrieval respectively. These simple baselines yield better results than state-of-the-art methods, both qualitatively and quantitatively. We show that encoder-decoder methods are statistically indistinguishable from these baselines, thus indicating that the current state of the art in single-view object reconstruction does not actually perform reconstruction but image classification. We identify aspects of popular experimental procedures that elicit this behavior and discuss ways to improve the current state of research.
引用
收藏
页码:3400 / 3409
页数:10
相关论文
共 50 条
  • [1] Single-View 3D Reconstruction of Curves
    Fakih, Ali
    Wilser, Nicola
    Maillot, Yvan
    Cordier, Frederic
    ADVANCES IN COMPUTER GRAPHICS, CGI 2023, PT II, 2024, 14496 : 3 - 14
  • [2] Learning View Priors for Single-view 3D Reconstruction
    Kato, Hiroharu
    Harada, Tatsuya
    2019 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2019), 2019, : 9770 - 9779
  • [3] What's on the Other Side? A Single-View 3D Scene Reconstruction
    Staszak, Rafal
    Kulecki, Bartlomiej
    Sempruch, Witold
    Belter, Dominik
    2022 17TH INTERNATIONAL CONFERENCE ON CONTROL, AUTOMATION, ROBOTICS AND VISION (ICARCV), 2022, : 173 - 180
  • [4] Photometric single-view dense 3D reconstruction in endoscopy
    Batlle, Victor M.
    Montiel, J. M. M.
    Tardos, Juan D.
    2022 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS (IROS), 2022, : 4904 - 4910
  • [5] Single-view 3D reconstruction via dual attention
    Li, Chenghuan
    Xiao, Meihua
    Li, Zehuan
    Chen, Fangping
    Wang, Dingli
    PEERJ COMPUTER SCIENCE, 2024, 10
  • [6] Domain-Adaptive Single-View 3D Reconstruction
    Pinheiro, Pedro O.
    Rostamzadeh, Negar
    Ahn, Sungjin
    2019 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2019), 2019, : 7637 - 7646
  • [7] Dynamic Domain Adaptation for Single-view 3D Reconstruction
    Yang, Cong
    Xie, Housen
    Tian, Haihong
    Yu, Yuanlong
    2021 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS (IROS), 2021, : 3563 - 3570
  • [8] A Dataset-Dispersion Perspective on Reconstruction Versus Recognition in Single-View 3D Reconstruction Networks
    Zhou, Yefan
    Shen, Yiru
    Yan, Yujun
    Feng, Chen
    Yang, Yaoqing
    2021 INTERNATIONAL CONFERENCE ON 3D VISION (3DV 2021), 2021, : 1331 - 1340
  • [9] Single-view 3D Reconstruction Algorithm Based on View-aware
    Wang Nian
    Hu Xuyang
    Zhu Fan
    Tang Jun
    JOURNAL OF ELECTRONICS & INFORMATION TECHNOLOGY, 2020, 42 (12) : 3053 - 3060
  • [10] Single-View 3D Mesh Reconstruction for Seen and Unseen Categories
    Yang, Xianghui
    Lin, Guosheng
    Zhou, Luping
    IEEE TRANSACTIONS ON IMAGE PROCESSING, 2023, 32 : 3746 - 3758