PolyFit reconstruction pipeline
PolyFit implements the hypothesis and selection based surface reconstruction method described in the following paper:
Liangliang Nan and Peter Wonka.
PolyFit: Polygonal Surface Reconstruction from Point Clouds.
ICCV 2017.
Please consider citing the above paper if you use the code/program (or part of it).
Prebuilt executable files (for both macOS and Windows) are available here.
You can also build PolyFit from the source code:
-
Download the source code.
-
Dependencies
-
Build PolyFit
- PolyFit.sln: for Microsoft Visual Studio (tested on 64bits Windows 10)
- PolyFit.pro: QtCreator (for macOS, Linux, Windows, etc.)
Super easy! This demo version provides a user interface with a few buttons (with numbered icons) and screen hints corresponding to these steps. Just click the buttons following the hints.
Some test data can be downloaded from the project page.
More infomation about the data (e.g., data format) are described here.
Incorporating plane extraction adds an unnecessary dependency to more third-party libraries (e.g., RANSAC). Besides, it has some randomness (due to the nature of RANSAC) and the data quality can vary a lot (it should be fine if some regions of the planes are missing). So I isolated this part from this demo version and you're expected to provide the planar segments as input.
You can use my Mapple to extract planes from point clouds. After you load the point cloud, go to the menu Partition -> Extract Primitives. To visualize the planes, change the renderer from 'Plain' to 'Group' in the Rendering panel (at the left side of Mapple). You can save the planes as bvg (Binary Vertex Group) format. The ASCII format vg also works but slow.
Four solvers, namely Gurobi, SCIP, GLPK, and lp_solve, are provided (with source code) in PolyFit. The Gurobi solver is more efficient and reliable and should always be your first choice. In case you want a fast but open source solver, please try SCIP, which is slower than Gurobi but acceptable. The GLPK and lp_solve solvers only manage to solve small problems. They are too slow (and may not guarantee to succeed). For example the data "Fig1", Gurobi takes only 0.02 seconds, while lp_solve 15 minutes. For your convenience, the dynamic library of Gurobi is included in this distribution, but you may still need to obtain a license (free for academic use) from here.
This demo implementation incorporates a progress logger in the user interface. Thus, running times should be (slightly) longer than those reported in our paper.
This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3 of the License or (at your option) any later version. The full text of the license can be found in the accompanying LICENSE file.
Should you have any questions, comments, or suggestions, please contact me at: liangliang.nan@gmail.com
Liangliang Nan
https://3d.bk.tudelft.nl/liangliang/
July 18, 2017
Copyright (C) 2017