Polarimetric Multi-View 3D Reconstruction

Okutomi-Monno Lab.
Institute of Science Tokyo


Introduction of Polarimetric Multi-View 3D Reconstruction

Polarimetric Multi-View 3D Reconstruction aims at recovering detailed 3D shape from multi-view polarization images under unconstrained conditions. Previous multi-view 3D reconstruction methods using RGB images are inherently weak in texture-less regions without enough photometric information. To solve this problem, we introduce polarimetric information into the process of reconstruction which indicates strong relationship with the scene geometry. Our proposed polarimetric multi-view 3D reconstruction methods handle polarimetric ambiguities and integrate polarimetric information efficiently with RGB information to achieve better results than existing methods.

Simple Yet Effective Way to Use Polarimetric Information in Stereo Matching [Project Page]

Fig2
Our proposed architecture for polarimetric stereo matching using RGB and AoP images.

Publication:
Simple Yet Effective Way to Use Polarimetric Information in Stereo Matching
Jinyu Zhao, Yusuke Monno, and Masatoshi Okutomi
International Conference on Machine Vision Applications (MVA), 2025.

Polarimetric PatchMatch Multi-View Stereo [Project Page]

Fig3
The flowchart of PolarPMS using multi-view RGB and AoP images.

Publication:
Polarimetric PatchMatch Multi-View Stereo
Jinyu Zhao, Jumpei Oishi, Yusuke Monno, and Masatoshi Okutomi
IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pp.3464-3472, 2024.

Polarimetric Multi-View Inverse Rendering (Polarimetric MVIR) [Project Page]

Fig4
The flowchart of Polarimetric MVIR using multi-view RGB and AoP images.

Publication:
Polarimetric Multi-View Inverse Rendering
Jinyu Zhao, Yusuke Monno, Masatoshi Okutomi
IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol.45, Issue 7, pp.8798-8812, 2023.

アクセスカウンター