Point Registration via Efficient Convex Relaxation

SIGGRAPH 2016


Reviews

Information

  • Paper topic: Geometry
  • Software type: Code
  • Able to run a replicability test: True
  • Replicability score: 5
  • Software language: Matlab / Mathematica / ..
  • License: unspecified
  • Build mechanism: Not applicable (python, Matlab..)
  • Dependencies: matlab
  • Documentation score {0,1,2}: 2
  • Reviewer: Nicolas Bonneel <nicolas.bonneel@liris.cnrs.fr>
  • Time spent for the test (build->first run, timeout at 100min): 60min

Source code information

Comments

A few steps to make the program run on Windows+Matlab R2018a:
- in all test files, replace
addpath(genpath('~/mosek'))
addpath(genpath('~/yalmip'))
by the actual paths, e.g., 
addpath(genpath('C:\Program Files\Mosek\9.1\toolbox\R2015a'))
addpath(genpath('C:\libs\YALMIP-master'))

and similarly for the toolbox_graph (downloadable at https://fr.mathworks.com/matlabcentral/fileexchange/5355-toolbox-graph ) or the FAUST data (e.g., addpath('C:\ICCV_2015_source_code\training\watertight_5000') )

- in interleaving.m, line 82, replace
 X_proj = reshape( intlinprog(fObj,intcon,[],[],A_ds,b_ds,lb,ub,options ), n, n );
 by
  X_proj = reshape( intlinprog(fObj,intcon,[],[],A_ds,b_ds,lb,ub,struct(options) ), n, n );

 - in interleaving.m, line 97, replace 
 X_proj = reshape( intlinprog(fObj,intcon,[],[],A_ds,b_ds,lb,ub,options ), n, n );
 by
 X_proj = reshape( intlinprog(fObj,intcon,[],[],A_ds,b_ds,lb,ub,struct(options) ), n, n );

While this produces a warning, this prevents an error in Mosek's intlinprog at line 135
  options          = mskoptimset(defaultopt,options)
  
- Make sure you have about 10GB of free RAM, otherwise Mosek crashes.
- If you installed Mosek after having installed Yalmip and already run the code unsuccessfully, make sure to "clear CACHED_SOLVERS" otherwise Mosek won't be found by Yalmip as the available solvers are cached.
 
The Faust dataset is at http://faust.is.tue.mpg.de/ (though I don't think the original dataset is needed) and the iccv code has moved from web.stanford.edu/~cqf/convex to   https://cqf.io/convex/

The different tests for 'synthetic', 'faust' and 'scape' run smoothly and produce the expected matchings. I did not test the last script, testPMSDP_scapeRawScans, since it requires manually preprocessing the SCAPE dataset with the ICCV code.

Another run on Linux produced similar results.

If you want to contribute with another review, please follow these instructions.

Please consider to cut/paste/edit the raw JSON data attached to this paper.