tsdf-fusion-python

Why we included this project

TSDF fusion is the classic step that folds a stack of registered RGB-D frames into a single voxel volume, and this project keeps that operation compact enough to read in one sitting. Give it aligned color and depth images and it builds a projective truncated signed distance function volume, then exports a surface mesh or point cloud. A pure CPU path runs on any machine, while an optional PyCUDA backend runs the same workload far faster on NVIDIA hardware, so you can prototype without a GPU and speed things up when one is available. The code is short and legible, which makes it a handy reference for the KinectFusion-style math behind a lot of reconstruction work in the literature, and it has been used in published projects such as 3DMatch.