plan2scene

Why we included this project

Plan2Scene is the official implementation of a CVPR 2021 paper that converts a 2D floorplan and a handful of room photos into a textured 3D mesh of a residence. The pipeline lifts the floorplan into a mesh, synthesizes tileable textures for the dominant surfaces, and uses a graph neural network to infer textures for areas the photos never captured. That last step is the interesting part: it handles sparse, unaligned photo coverage rather than assuming a complete scan. The repo ships pretrained models and a Colab demo, with setup instructions in the README, though it depends on an external texture-synthesis binary and the Rent3D++ dataset, so expect some assembly before anything runs. It is research code aimed at replicating the paper's results, not a drop-in production tool, but teams building floorplan-to-3D or interior texturing systems will find the approach worth studying.