Get 3D shape of object by neural reconstruction by derivatives and coarse-to-fine with Neuralangelo
Get 3D shape of object by neural reconstruction by derivatives and coarse-to-fine with Neuralangelo
Neuralangelo: High-Fidelity Neural Surface Reconstruction
arXiv paper abstract https://arxiv.org/abs/2306.03092
arXiv PDF paper https://arxiv.org/pdf/2306.03092.pdf
Neural surface reconstruction has been shown to be powerful for recovering dense 3D surfaces via image-based neural rendering.
However, current methods struggle to recover detailed structures of real-world scenes.
… present Neuralangelo, which combines the representation power of multi-resolution 3D hash grids with neural surface rendering.
Two key ingredients enable … approach: (1) numerical gradients for computing higher-order derivatives as a smoothing operation and (2) coarse-to-fine optimization on the hash grids controlling different levels of details.
… without auxiliary inputs such as depth, Neuralangelo can … recover … 3D surface … from multi-view images with fidelity significantly surpassing previous methods, enabling detailed large-scale scene reconstruction from RGB video captures.
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
