merge develop. 0.5 release.

.gitignore

@@ -4,3 +4,8 @@
 *.toc
 *.bbl
 *.blg
+*.cbp*
+*.layout*
+*~
+*.pyc
+*.db

.gitmodules

@@ -4,3 +4,6 @@
 [submodule "src/dependencies/OpenExif"]
 	path = src/dependencies/OpenExif
 	url = https://github.com/openMVG-thirdparty/OpenExif.git
+[submodule "src/dependencies/osi_clp"]
+	path = src/dependencies/osi_clp
+	url = https://github.com/openMVG-thirdparty/osi_clp.git

.travis.yml

@@ -0,0 +1,21 @@
+language: cpp
+
+compiler:
+  - gcc
+
+install: sudo apt-get install cmake libpng-dev libjpeg8-dev libxxf86vm1 libxxf86vm1 libxxf86vm-dev x11proto-xf86vidmode-dev libxrandr-dev
+
+before_script:
+  - cd ..
+  - mkdir build
+  - cd build
+  - cmake -DCMAKE_BUILD_TYPE=release . ../openMVG/src
+
+script:
+  - make
+  - make test
+
+branches:
+  only:
+    - develop
+

AUTHORS

@@ -7,6 +7,7 @@ Renaud Marlet <[email protected]>
 ySalaun
 bduisit
 ORNis
+fcastan
 
 -------------
 OpenMVG authors would thanks the libmv authors

BUILD

@@ -36,6 +36,15 @@ Setup the required external library.
  $ cd openMVG_Build
  $ cmake -DCMAKE_BUILD_TYPE=RELEASE . ../openMVG/src/
 
+=> In order to use the MOSEK backend for the linear programming oepnMVG module
+  - Check that you have an uptodate MOSEK licence, else openMVG MOSEK unit test will fail.
+
+ $ cmake -DCMAKE_BUILD_TYPE=RELEASE
+    -DMOSEK_SEARCH_HEADER="~/Documents/Lib/mosek/6/tools/platform/linux64x86/h"
+    -DMOSEK_SEARCH_LIB="~/Documents/Lib/mosek/6/tools/platform/linux64x86/bin"
+    . ../openMVG/src/
+
+
 If you want have an IDE openable project with codeblocks:
  $ cmake -G "CodeBlocks - Unix Makefiles" -DCMAKE_BUILD_TYPE=RELEASE . ../openMVG/src/
 

README.md

@@ -20,9 +20,9 @@ Building
 
 See [BUILD](https://github.com/openMVG/openMVG/raw/master/BUILD) text file
 
-Continuous integration (master, develop branch, linux 64 bits):
+Continuous integration (develop branch, linux 64 bits, BUILD + UNIT TESTING):
 
-[![Build Status](https://drone.io/github.com/openMVG/openMVG/status.png)](https://drone.io/github.com/openMVG/openMVG/latest)
+[![Build Status](https://travis-ci.org/openMVG/openMVG.png?branch=develop)](https://travis-ci.org/openMVG/openMVG)
 
 -------
 License

docs/sphinx/rst/conf.py

@@ -59,9 +59,9 @@
 # built documents.
 #
 # The short X.Y version.
-version = '0.4'
+version = '0.5'
 # The full version, including alpha/beta/rc tags.
-release = '0.4'
+release = '0.5'
 
 # The language for content autogenerated by Sphinx. Refer to documentation
 # for a list of supported languages.

docs/sphinx/rst/dependencies/glfw.rst

@@ -8,9 +8,9 @@ USAGE
 
 openMVG uses GLFW library in order to provide an easy way to create openGL windows context.
 
-==========
+===========
 Description
-==========
+===========
 
 "[GLFW]_ is an Open Source, multi-platform library for creating windows with OpenGL contexts and managing input and events.
 It is easy to integrate into existing applications and does not lay claim to the main loop.

docs/sphinx/rst/dependencies/openExif.rst

@@ -8,9 +8,9 @@ USAGE
 
 openMVG uses OpenEXIF library in order to read JPEG EXIF metadata.
 
-==========
+===========
 Description
-==========
+===========
 
 "[OpenExif]_ is an object-oriented library for accessing Exif formatted JPEG image files. The toolkits allow for creating, reading, and modifying the metadata in the Exif file. It also provides mean of getting and setting the main image and the thumbnail image.
 "

docs/sphinx/rst/dependencies/osi_clp.rst

@@ -10,11 +10,11 @@ openMVG uses the [OSI]_ and the [CLP]_ solver in order to solve linear programs
 
 CLP has been choosen because it is known to support problems of up to 1.5 million constraints [CLP FAQ]_.
 
-[LPSOLVE]_ have been tested but tests shown that it is less reliable (sometimes, there is no convergence to a solution).
+[LPSOLVE]_ has been tested but tests shown that it is less reliable (sometimes, there is no convergence to a solution).
 
-==========
+===========
 Description
-==========
+===========
 
 [OSI]_
   Osi (Open Solver Interface) provides an abstract base class to a generic linear programming (LP) solver, along with derived classes for specific solvers. Many applications may be able to use the Osi to insulate themselves from a specific LP solver. (CLP and MOSEK backend are supported in openMVG).
@@ -25,5 +25,5 @@ Description
 .. [LP] http://en.wikipedia.org/wiki/Linear_programming
 .. [OSI] https://projects.coin-or.org/Osi
 .. [CLP] https://projects.coin-or.org/Clp/wiki
-.. [CLP FAQ] https://projects.coin-or.org/Clp/wiki/FAQ
+.. [CLP_FAQ] https://projects.coin-or.org/Clp/wiki/FAQ
 .. [LPSOLVE] http://lpsolve.sourceforge.net/5.5/

docs/sphinx/rst/index.rst

@@ -10,7 +10,7 @@ Introduction
 ==================
 OpenMVG (Multiple View Geometry) is a library for computer-vision scientists and especially
 targeted to the Multiple View Geometry community. It is designed to provide an easy access to the
-classical problem solvers in Multiple View Geometry and solve them accurately..
+classical problem solvers in Multiple View Geometry and solve them accurately.
 
 ############################
 Why another library
@@ -28,7 +28,7 @@ openMVG library overview
 
 The openMVG library is cut in various modules:
 
-* **Libraries**,
+* **Libraries, core**,
 
   * comes with unit tests that assert algorithms results and show how use the code.
 
@@ -38,7 +38,11 @@ The openMVG library is cut in various modules:
 
 * **Binaries**,
 
-  * softwares build to perform toolchain processing (features matching, Structure from Motion ...).
+  * softwares build to perform toolchain processing
+
+    * features matching in un-ordered photo collection,
+    * SfM: Structure from Motion,
+    * color harmonization of photo collection.
 
 Contents:
 
@@ -47,21 +51,20 @@ Contents:
 
    openMVG/openMVG.rst
    openMVG_Samples/openMVG_Samples.rst
+   software/softwares.rst
    patented/patented.rst
-   software/SfM/SfM.rst
-   testing/testing.rst
    dependencies/external_libs.rst
    third_party/third_party.rst
    FAQ/FAQ.rst
 
 
-
 ############################
 Acknowledgements
 ############################
-openMVG authors would like to thank libmv authors for providing an inspiring base to design the
-openMVG library. Authors also would like to thank Mikros Image and LIGM-Imagine laboratory
-for support and authorization to make this library as an open-source project.
+openMVG authors would like to thank:
+
+- libmv authors for providing an inspiring base to design the openMVG library.
+- Mikros Image and LIGM-Imagine laboratory for support and authorization to make this library as an open-source project.
 
 ############################
 License
@@ -74,8 +77,8 @@ Please refer to the license file contained in the source for complete license de
 Dependencies
 ############################
 OpenMVG come as a standalone distribution, you don't need to install libraries to make it compiles
-and run. Exception for the Linux library is made for the png, zlib and jpeg library in order to use
-the one of the system.
+and run.
+On Linux the library will use if available the local png, zlib and jpeg libraries.
 
 Indices and tables
 ==================

docs/sphinx/rst/openMVG/linear_programming/linfinityCV.rst

@@ -0,0 +1,36 @@
+########################################################
+L infinity solvers for computer vision
+########################################################
+
+openMVG propose Linear programming based solver for various problem in computer vision by minimizing of the maximal error the residuals between observations and estimated parameters (The L infinity norm).
+
+openMVG implements problems introduced by [LinfNorm]_ and generalized by [LinfNormGeneric].
+
+Rather than considering quadratic constraints that require SOCP (Second Orde Cone Programming) we consider their LP (linear program) equivalent. It makes usage of residual error expressed with absolute error ( ``|a|<b`` inequality is transformed in two linear inequalities ``a<b`` and ``-b<-a``. It makes the solving faster and constraint easier to express (see. [Arnak]_ for more explanation).
+
+- N-view triangulation [LinfNorm]_,
+- Resection or pose matrix estimation [LinfNorm]_,
+- Estimation of translations and structure from known rotations,
+
+  - two formulation are implemented,
+
+    - the simple one [LinfNorm]_,
+    - the robust based on slack variables [OlssonDuality]_.
+
+- Registration of relative translations to compute global translations [ICCV13]_,
+
+  - using triplets of translations.
+
+
+
+.. [LinfNorm] L infinity Minimization in Geometric Reconstruction Problems. Richard I. Hartley, Frederik Schaffalitzky. CVPR 2004.
+
+.. [LinfNormGeneric] "Multiple-View Geometry under the L infty Norm." Authors: Fredrik Kahl, Richard Hartley. 2008. and "Multiple View Geometry and the L infty -norm". Fredrik Kahl. ICCV 2005.
+
+.. [Arnak] Robust estimation for an inverse problem arising in multiview geometry. Arnak S. Dalalyan, Renaud Keriven. In J. Math. Imaging Vision, 2012.
+
+
+.. [OlssonDuality] Outlier Removal Using Duality. Carl Olsson, Anders Eriksson and Richard Hartley, Richard. CVPR 2010.
+
+
+.. [ICCV13] Global Fusion of Relative Motions for Robust, Accurate and Scalable Structure from Motion. Pierre Moulon, Pascal Monasse and Renaud Marlet. ICCV 2013.

docs/sphinx/rst/openMVG/linear_programming/lp.rst

@@ -0,0 +1,153 @@
+*************************
+linear programming
+*************************
+
+Linear programming [LP]_ is a technique for the optimization of a linear objective function, subject to linear equality and linear inequality constraints such as:
+
+.. math::
+
+    \begin{align} & \text{maximize} && \mathbf{c}^\mathrm{T} \mathbf{x}\\
+    & \text{subject to} && A \mathbf{x} \leq \mathbf{b} \\
+    & \text{and} && \mathbf{x} \ge \mathbf{0} \end{align}
+
+where ``x`` represents the vector of variables (to be determined), ``c`` and ``b`` are vectors of (known) coefficients, ``A`` is a (known) matrix of coefficients.
+
+openMVG linear programming tools
+---------------------------------
+
+openMVG provides tools to:
+
+- configure Linear programs (LP container),
+- solve Linear Programs (convex or quasi convex ones).
+
+It results in a collection of L infinity based solver for computer vision problems.
+
+.. toctree::
+  :maxdepth: 1
+
+  linfinityCV.rst
+
+
+openMVG linear program container
+---------------------------------
+
+openMVG provides a generic container for LP (Linear Programming problems) that can be dense of sparse.
+
+.. code-block:: c++
+
+  // Dense LP
+  LP_Constraints
+  // Sparse LP
+  LP_Constraints_Sparse
+
+
+It allows to embed:
+
+- objective function ``c`` and the problem type (minimization or maximization),
+- constraints (coefficients ``A``, Sign, objective value ``b``),
+- bounds over ``x`` parameters (<=, =, >=).
+
+openMVG linear program solvers
+---------------------------------
+
+openMVG provide access to different solvers (not exhaustive):
+
+- OSI_CLP (COIN-OR) project,
+- MOSEK commercial, free in a research context.
+
+Those solver have been choosen due to the stability of their results and ability to handle large problems without numerical stability (LPSolve and GPLK have been discarded after extensive experiments).
+
+I refer the reader to openMVG/src/openMVG/linearProgramming/linear_programming_test.cpp to know more.
+
+openMVG linear programming module usage
+-------------------------------------------
+
+The linear programming module of openMVG can be used for:
+
+- solve classical linear problem (optimization),
+- test the feasibility of linear problem,
+- optimize upper bound of feasible problem (quasi-convex linear programs).
+
+**classical linear problem solving (optimization)**
+
+Here an example of usage of the framework:
+
+.. code-block:: c++
+
+  // Setup the LP (fill A,b,c and the constraint over x)
+  LP_Constraints cstraint;
+  BuildLinearProblem(cstraint);
+
+  // Solve the LP with the solver of your choice
+  std::vector<double> vec_solution(2);
+  #if OPENMVG_HAVE_MOSEK  
+    MOSEK_SolveWrapper solver(2);
+  #else
+    OSI_CLP_SolverWrapper solver(2);
+  #endif
+  // Send constraint to the LP solver
+  solver.setup(cstraint);
+
+  // If LP have a solution
+  if (solver.solve())
+    // Get back estimated parameters
+    solver.getSolution(vec_solution);
+
+**Linear programming, feasible problem**
+
+openMVG can be use also to test only the feasibility of a given LP
+
+
+.. math::
+
+    \begin{align} & \text{find} && \mathbf{x}\\
+    & \text{subject to} && A \mathbf{x} \leq \mathbf{b} \\
+    & \text{and} && \mathbf{x} \ge \mathbf{0} \end{align}
+
+**Linear programming, quasi convex optimization**
+
+openMVG used a lot of L infinity minimisation formulation.
+Often the posed problems are quasi-convex and dependent of an external parameter that we are looking for (i.e the maximal re-projection error for which the set of contraint is still feasible).
+
+
+Optimization of this upper bound parameter can be done by iterating over all the possible value or by using a bisection that reduce the search range at each iteration.
+
+.. code-block:: c++
+
+  Require: gammaLow, gammUp (Low and upper bound of the parameter to optimize)
+  Require: the LP problem (cstraintBuilder)
+  Ensure: the optimal gamma value, or return infeasibility of the contraints set.
+
+  BisectionLP(
+    LP_Solver & solver,
+    ConstraintBuilder & cstraintBuilder,
+    double gammaUp  = 1.0,  // Upper bound
+    double gammaLow = 0.0,  // lower bound
+    double eps      = 1e-8, // precision that stop dichotomy
+    const int maxIteration = 20) // max number of iteration
+  {
+    ConstraintType constraint;
+    do
+    {
+      ++k; // One more iteration
+
+      double gamma = (gammaLow + gammaUp) / 2.0;
+
+      //-- Setup constraint and solver
+      cstraintBuilder.Build(gamma, constraint);
+      solver.setup( constraint );
+
+      //-- Solving
+      bool bFeasible = solver.solve();
+
+      //-- According feasibility update the corresponding bound
+      //-> Feasible, update the upper bound
+      //-> Not feasible, update the lower bound
+      (bFeasible) ? gammaUp = gamma; : gammaLow = gamma;
+
+    } while (k < maxIteration && gammaUp - gammaLow > eps);
+  }
+
+
+
+.. [LP] http://en.wikipedia.org/wiki/Linear_programming