Get 3D shape of object by combining neural reconstruction and multiple views with C2F2NeUS
Get 3D shape of object by combining neural reconstruction and multiple views with C2F2NeUS
C2F2NeUS: Cascade Cost Frustum Fusion for High Fidelity and Generalizable Neural Surface Reconstruction
arXiv paper abstract https://arxiv.org/abs/2306.10003
arXiv PDF paper https://arxiv.org/pdf/2306.10003.pdf
There is … effort to combine … multi-view stereo (MVS) and neural implicit surface (NIS), in scene reconstruction from sparse views.
… introduce a novel integration scheme that combines the multi-view stereo with neural signed distance function representations, which potentially overcomes the limitations of both methods.
MVS uses per-view depth estimation and cross-view fusion to generate accurate surface, while NIS relies on a common coordinate volume.
… propose to construct per-view cost frustum for finer geometry estimation, and then fuse cross-view frustums and estimate the implicit signed distance functions to tackle noise and hole issues.
… apply a cascade frustum fusion strategy to effectively captures global-local information and structural consistency.
… method reconstructs robust surfaces and outperforms existing state-of-the-art methods.
Stay up to date. Subscribe to my posts https://morrislee1234.wixsite.com/website/contact
Web site with my other posts by category https://morrislee1234.wixsite.com/website
LinkedIn https://www.linkedin.com/in/morris-lee-47877b7b
