Learning Signed Distance Field for Multi-view Surface Reconstruction

被引:42
|
作者
Zhang, Jingyang [1 ]
Yao, Yao [1 ]
Quan, Long [1 ]
机构
[1] Hong Kong Univ Sci & Technol, Hong Kong, Peoples R China
关键词
STEREO;
D O I
10.1109/ICCV48922.2021.00646
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Recent works on implicit neural representations have shown promising results for multi-view surface reconstruction. However, most approaches are limited to relatively simple geometries and usually require clean object masks for reconstructing complex and concave objects. In this work, we introduce a novel neural surface reconstruction framework that leverages the knowledge of stereo matching and feature consistency to optimize the implicit surface representation. More specifically, we apply a signed distance field (SDF) and a surface light field to represent the scene geometry and appearance respectively. The SDF is directly supervised by geometry from stereo matching, and is refined by optimizing the multi-view feature consistency and the fidelity of rendered images. Our method is able to improve the robustness of geometry estimation and support reconstruction of complex scene topologies. Extensive experiments have been conducted on DTU, EPFL and Tanks and Temples datasets. Compared to previous state-of-the-art methods, our method achieves better mesh reconstruction in wide open scenes without masks as input.
引用
收藏
页码:6505 / 6514
页数:10
相关论文
共 50 条
  • [1] Multi-View Reconstruction using Signed Ray Distance Functions (SRDF)
    Zins, Pierre
    Xu, Yuanlu
    Boyer, Edmond
    Wuhrer, Stefanie
    Tung, Tony
    2023 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2023, : 16696 - 16706
  • [2] VolRecon: Volume Rendering of Signed Ray Distance Functions for Generalizable Multi-View Reconstruction
    Ren, Yufan
    Wang, Fangjinhua
    Zhang, Tong
    Pollefeys, Marc
    Süsstrunk, Sabine
    2023 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2023, : 16685 - 16695
  • [3] Multi-view surface reconstruction using polarization
    Atkinson, GA
    Hancock, ER
    TENTH IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION, VOLS 1 AND 2, PROCEEDINGS, 2005, : 309 - 316
  • [4] NeuralUDF: Learning Unsigned Distance Fields for Multi-view Reconstruction of Surfaces with Arbitrary Topologies
    Long, Xiaoxiao
    Lin, Cheng
    Liu, Lingjie
    Liu, Yuan
    Wang, Peng
    Theobalt, Christian
    Komura, Taku
    Wang, Wenping
    2023 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2023, : 20834 - 20843
  • [5] SurRF: Unsupervised Multi-View Stereopsis by Learning Surface Radiance Field
    Zhang, Jinzhi
    Ji, Mengqi
    Wang, Guangyu
    Xue, Zhiwei
    Wang, Shengjin
    Fang, Lu
    IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 2022, 44 (11) : 7912 - 7927
  • [6] Multi-scale surface reconstruction based on a curvature-adaptive signed distance field
    Tang, Yizhi
    Feng, Jieqing
    COMPUTERS & GRAPHICS-UK, 2018, 70 : 28 - 38
  • [7] NeRSemble: Multi-view Radiance Field Reconstruction of Human Heads
    Kirschstein, Tobias
    Qian, Shenhan
    Giebenhain, Simon
    Walter, Tim
    Niessner, Matthias
    ACM TRANSACTIONS ON GRAPHICS, 2023, 42 (04):
  • [8] Sphere Packing Aided Surface Reconstruction for Multi-view Data
    Liu, Kun
    Galindo, Patricio A.
    Zayer, Rhaleb
    ADVANCES IN VISUAL COMPUTING (ISVC 2014), PT II, 2014, 8888 : 173 - 184
  • [9] Multi-view representation learning for multi-view action recognition
    Hao, Tong
    Wu, Dan
    Wang, Qian
    Sun, Jin-Sheng
    JOURNAL OF VISUAL COMMUNICATION AND IMAGE REPRESENTATION, 2017, 48 : 453 - 460
  • [10] Volume Sweeping: Learning Photoconsistency for Multi-View Shape Reconstruction
    Leroy, Vincent
    Franco, Jean-Sebastien
    Boyer, Edmond
    INTERNATIONAL JOURNAL OF COMPUTER VISION, 2021, 129 (02) : 284 - 299