neural-rgbd-surface-reconstruction

Why we included this project

Neural surface reconstruction from RGB-D video is a fast-moving area, and this repository is the implementation behind a CVPR 2022 paper that combines implicit surface representations with neural radiance fields. Feed it an RGB-D image sequence and it trains a multi-layer perceptron that outputs a watertight surface mesh, which makes it a natural reference point if you want to reproduce the paper's results or build on them. The codebase descends from the original NeRF pipeline, so anyone who already knows that code can get oriented quickly, and the repo documents the expected data layout and provides a dataset and evaluation meshes to test against. It is research software tied to a specific paper, so treat it as a baseline and a learning resource rather than something to drop into production, but for experiments in neural surface reconstruction it is a solid place to start.