This repository contains the implementation of the work "Partial Correspondence of 3D Shapes using Properties of the Nearest-Neighbor Field" by Nadav Y. Arbel, Ayellet Tal and Lihi-Zelnik Manor. Video
The movement from sparse to dense correspondence has been done by adjusting to code of [1] which resides here to accept sparse correspondence
Made in windows 10
- PCL(1.8.1 64 bit MSVC 2017)
- Might have to add VTK_DIR to environment variables : <PCL_intallation_folder>\3rdParty\VTK\lib\cmake\vtk7.0
- CMAKE
- matlab
- Visual Studio 2017 64 bit
- Open cmake
- set source code directory to: <this_repository_root>\nnf_sparse_code\
- set build dir to <source_dir>\build\
- configure (with Visual Studio 15 2017 Win64) and generate
- click on open project
- in Visual Studio make sure branch is set to release and build 3DDIS_OMP
You now should be able to run everything in the root dir
We provide instructions to replicate our results. Sparse correspondences reside in "<match_dir>\refined_sparse", while dense correspondences reside in "<match_dir>\dense"
To run this experiment please download the test set and evaluation code from their site.
Run runSHREC16A - just set the correct paths to the dataset inside the file.
For ground truth correspondences please contact the competition organizers, they had kindly provided me with the data, but had I have no permission to further divulge it myself.
Alternatively our obtained correspondences can be downloaded below.
To run this experiment please download the low resolution test set and evaluation code from their site
Run runSHREC16B - just set the correct paths to the dataset inside the file.
Alternatively our obtained correspondences can be downloaded below.
Here you need to provide geoedesic distance matrices for each model. I provide a matlab code which achieves this.
Download the dataset from the site.
Since we require sampling to make the problem tractable, and the dense correspondence algorithm[1] requires 2-manifold meshes of a specific area, you should copy the code from here into the model directory and run "run.bat"
Then run Run run_FAUST_test - just set the correct paths inside the file.
We have only made a qualitative evaluation here, as it contains full to part problems
to visualizing correspondences we provide an example script for SHREC'16:Partial[1] one code is for sparse correspondences, and another for dense correspondences