Mask3D

Why we included this project

Mask3D is a research implementation of a mask transformer for 3D semantic instance segmentation, the task of finding and labeling each distinct object in a point cloud rather than classifying the whole scene. Built on PyTorch with Lightning and Hydra, it achieves state-of-the-art results on the ScanNet, ScanNet200, S3DIS, and STPLS3D benchmarks, and the repo ships trained checkpoints and preprocessing scripts for each dataset, so reproducing the reported numbers takes little extra work. The code is research-oriented, so expect to adapt it to your own data and hardware before it fits into a production pipeline. For teams working on indoor scene understanding, robotics perception, or urban point-cloud analysis, it is a useful reference for separating individual instances, and the modular structure makes it a reasonable starting point for extending the work.