Difference between revisions of "User:NyahCheck/Survey of CSG Algorithms"

From BRL-CAD
(Project Summary)
(Contributions)
 
(13 intermediate revisions by the same user not shown)
Line 5: Line 5:
 
= Personal Information =
 
= Personal Information =
  
Name: Nyah Watad Check
+
Name: Nyah Watad Check.
Email: check.nyah@gmail.com
+
Email: check.nyah@gmail.com.
IRC Nick: Ch3ck, Ch3ck_
+
IRC Nick: Ch3ck, Ch3ck_.
  
 
== Background Information ==
 
== Background Information ==
Line 19: Line 19:
  
 
* Google Summer of Code Participant, BRL-CAD, April - September 2013
 
* Google Summer of Code Participant, BRL-CAD, April - September 2013
**Implemented a pull routine to reverse the effects of a push on Geometry, **integrating command into MGED interface. (C, XML, 500+ lines of code)
+
**Implemented a pull routine to reverse the effects of a push on Geometry.
 +
**Integrated command into MGED interface. (C, XML, 500+ lines of code)
 
**Tested the polynomial and matrix math routines, improving the speeds of the inverse matrix and matrix determinant routines by over 10%. (C, 300+ lines of code)
 
**Tested the polynomial and matrix math routines, improving the speeds of the inverse matrix and matrix determinant routines by over 10%. (C, 300+ lines of code)
 
**Integrated the xpush routine which pushes objects having more than a single matrix transformation into the push which pushes object matrix transformations to leaf nodes. ( C, 1000 lines of code.)
 
**Integrated the xpush routine which pushes objects having more than a single matrix transformation into the push which pushes object matrix transformations to leaf nodes. ( C, 1000 lines of code.)
Line 29: Line 30:
 
**Languages: C(Excellent), Java( Excellent ), C++( Beginner), Bash(Excellent), SQL(Proficient)
 
**Languages: C(Excellent), Java( Excellent ), C++( Beginner), Bash(Excellent), SQL(Proficient)
 
**Tools: Secure Shell, subversion, Git, Linux, Netbeans, gdb, valgrind.
 
**Tools: Secure Shell, subversion, Git, Linux, Netbeans, gdb, valgrind.
 
 
  
 
= Synopsis/ Proejct Summary =
 
= Synopsis/ Proejct Summary =
Line 39: Line 38:
 
== Introduction ==
 
== Introduction ==
  
 +
Constructive Solid Geometry is a technique used in Solid modeling which allows a modeler to create a complex surface using Boolean operators to combine objects. This is often referred to as procedural modeling and can be performed on polygonal meshes. Here, the simplest solid objects used for representation are called primitives which are used to construct more complex objects using allowable Boolean operators such as union, intersection, difference and well as geometric transformations on those sets of objects.
 +
 +
OpenSCAD uses the Clipper Library to implement CSG in 2D and CGAL libraries to model 3D surfaces. My objective in this study is to look at the CSG Algorithms implemented by OpenSCAD and do a comparative study with other CSG algorithms, performing and algorithmic analysis and implement prototypes which can be useful in the OpenSCAD framework.
 +
This entails studying 4 main research papers together with various implementations of the algorithms associated with them and implementing the most efficient on the OpenSCAD platform.
 +
 +
 +
==CSG Algorithm for OpenSCAD(Current Implementation)==
 +
 +
OpenSCAD is a parametric CSG based modelling software; which means objects are manipulated by editing the CSG descriptions of objects being rendered. It employs two different mechanisms for processing CSG descriptions. It uses the OpenCSG library for visualization during editing and the CGAL library for export created a meshed BREP output. However, the OpenCSG Library uses the Goldfeather and  SCS  algorithms to evaluate the 2D images of a CSG object from the given perspective. This is beneficial  because of  its speed and ability to render complex images within a short time.
 +
 +
However, CGAL employs the Nef Polyhedra for evaluating boolean operations  on CSGs. It  is robust and is able to handle non manifold input but proves incredibly slow in practice and not very suitable for time critical  applications . There are also overflow  issues that arise with using large size polyhedra. The used of  BSP tree proves more efficient for CSG evaluations than CGAL's  Nef based polyhedra .
 +
Below describes  the use of BSP based algorithm for CSG evaluations than CGAL's Nef-based Polyhedra  system.
 +
 +
==BSP Tree based Implementation of Boolean Operations==
 +
 +
Bernstein et all[1] presented an exact and more efficient method for evaluating 3D polyhedra  already supported by OpenSCAD. Results from tests showed for a BSP tree implementation which 16-28X faster at performing iterative computations than CGAL's Nef polyhedra . The use of a BSP-tree based boolean algorithm allows the explicit handling of all geometric degeneracies without treating a large number of cases. There are also efficient  CSG to BSP algorithms which  could be used for easy representations in both formats.
 +
 +
BSP trees afford an alternative to B-rep algorithms that avoid their concomitant case explosion by explicitly handling all degenerate configurations of geometry. They have demonstrated to have suitable performance for interactive volumetric sculpting.
 +
 +
*Numeric Substrates
 +
In this system, a plane is a quadruple of floating point numbers interpreted as coefficients of a plane equation.
 +
**Concidence: Two planes p, q are coincident if and only if the determinant of all 2x2 minors are 0
 +
**coincident orientation:  if p and q are coincident and Pa.Qa, Pb.Qb, Pc.Qc, Pd.Qd are non negative.
 +
 +
**Point Validity: if a point A(p,q,r) is valid if i'ts determinant is non zero
 +
  Pa Pb Pc
 +
  Qa Qb Qc
 +
  Ra  Rb Rc
 +
 +
**Orientation: Given a point A(p, q, r) is valid and it lies behind, on, or in-front of the plane S if  and only iff the following expression is negative, zero or positive
 +
                            Pa Pb Pc  Pd
 +
Pa Pb Pc                    Qa Qb Qc Qd
 +
Qa Qb Qc      *            Ra Rb Rc Rd
 +
Ra Rb Rc                    Sa Sb Sc  Sd
 +
 +
These predicates are implemented as static filtered floating-point predicates in the style of shewchuk, but deviations will be done to support double precision.
 +
 +
*Geometric Substrates
 +
In the preceding step, we defined planes as primitives, points as triples of planes and 4 predicates operating on these plans; now we define a convex polygon type for a constructor and splitting routine.
 +
 +
**Convex polygon: This is a plane of support, s with a bounding plane {Bi} I £ Z. This vertices given by  Vi = (s, Bi-1, Bi).
 +
 +
**Construction of a Convex Polygon from a Plane: Given a plane h, this operation constructs a convex polygon representing h clipped by a very large box. The output polygon serves as a stand in for the infinite extent plane. A consistent axis aligned “very large box” is used for all calls to the constructor and consists of volume spaced bounded by planes X+, X-, Y+, y- , Z+, Z-.
 +
 +
**Splitting a Convex Polygon by a Plane: Since  polygon splitting may be viewed as two successive and complementarty instances of polygon clipping. The algorithm below similar to the Sutherland-Hodgman style polygon clippers, rather than deciding if and when to insert a crossing point to the output stream, we decide if and when to insert the splitting plane into the output stream.
 +
 +
==Algorithm ==
 +
If s is concident with h
 +
  if s is similarly oriented to h
 +
      Return (s, {Bi}i e Zn)
 +
  Else
 +
    Return nothing
 +
Else
 +
  For I e Zn(in order)
 +
      Output planes as specified by table lookup using:
 +
      o(s, bi-2, bi-1, h)
 +
      o(s, bi-1, bi, h);
 +
      o(s, bi, bi+1, h)
 +
    Return (s, output)
 +
                  Algorithm: Clip(s, {bi}i e Zn) by h
 +
 +
The Above algorithm works in interoperability between input and output that is between the Point-based polygon soup and the Plane-based Representation working with Boolean operations.
 +
 +
= Sample Algorithms/Links =
 +
 +
* Gilbert Bernstein and Don Fussell: Fast, Exact, Linear Booleans
 +
*Sebastian Steuer: Methods for Polygonalization of a Constructive Solid Geometry Description in Web-based Rendering Environments
 +
 +
* MEPP: Exact and Efficient Booleans for Polyhedra
 +
*Cork - by Gilbert Bernstein
 +
* CGAL: http://www.cgal.org/
 +
*http://liris.cnrs.fr/mepp/
 +
 +
=Testing and Verification=
  
= Links =
+
Test would be written to verify and test the robustness of the BSP-tree based algorithm for CSG evaluations. This could apply Octoball tests from single operations between sizable meshes to Random Boxes tests and Sculpt Bunny depending on the efficiency. Also, the heat sink test which has proven to have a worst case performance of O(n2) in the size of input can be applied to the aforementioned implementation. Unit tests will be developed to handle different aspects of the BSP-tree evaluations and appropriate documentation attached for later works on the system.
 +
Also, any bugs coming up during this process will be fixed and documentation of this work done during development since OpenSCAD is actively developed and this will keep other developers informed on any changes taking place in the CSG module for OpenSCAD.
  
= Deliverables =
+
=Deliverables=
 +
*A paper containing the CSG Algorithm survey and it's relation to OpenSCAD
 +
*Prototype implementations integrated with OpenSCAD on the more efficient and robust CSG Algorithms.
 +
*An Evaluation of how more advanced modeling techniques could be realised with each algorithm.
  
 
= Development Schedule/Timeline =
 
= Development Schedule/Timeline =
 +
This is a tentative plan which will be modified and developed as GSoC proceeds.
 +
 +
*May 17 – June 06( 3 weeks)
 +
**Study papers on this topic
 +
**Discuss with developers on implementation specifics and further clarifications on CSG Algorithms supported by OpenSCAD.
 +
 +
*May 24 – June 13 ( 3 weeks)
 +
**Check current implementation of the BSP tree algorithm for evaluating CSGs and porting to OpenSCAD.
 +
**Implement the BSP-tree representations  in OpenSCAD or optimize current implementation.
 +
**Testing, debugging and documentation.
 +
 +
*June 14  – June 20( 1 week)
 +
**Unit tests for BSP-tree routines
 +
**Testing degenerate cases.
 +
 +
*June 21 – July 4 ( 2 Weeks) Mid Term Evaluations
 +
**Add functionality for CSG to BSP-tree conversion
 +
**Testing for robustness and degenerate cases.
 +
 +
*July 5 – July 11( 1 week)
 +
**Implement testing suite for BSP tree implementation of the Bernstein algorithm.
 +
**Adding documentation to OpenSCAD framework
 +
 +
*July 12 – July 25 ( 2 weeks)
 +
 +
**Implement and test subroutines for Numeric substrates.
 +
**Testing and debugging.
 +
 +
**July 26 – August 8 ( 2 Weeks)
 +
**Add functionality for Geometric Substrates together with sub routines.
 +
**Testing and debugging of Convex polygon constructor for clipped plane
 +
 +
*August 9 – August 22( 2 Weeks)
 +
**Inside-out tests, address robustness issues with modules and improve performance of libraries.
 +
**Write paper on Comparison of CSG based algorithms in relation to OpenSCAD Implementations.
 +
**Check code for memory leaks and performance analysis, with unit tests added for implemented modules.
 +
 +
*August 23 – August 29( 2 weeks)
 +
**Pencils Down, Code clean up.
 +
**Review documentation and Finalize paper writing.
 +
**Final evaluation, Submission of code to melange.
  
 
= Time availability =
 
= Time availability =
 +
I would be able to offer over 40 hours on the project. However, since the project would start during our second Semester; I would be coding mostly during the nights up to late June or early July when our semester ends. Also, to meet up with the demands of the project, I would be coding during weekends  and regularly informing my mentors on the status of the project and regularly updating my logs in this respect.
  
 
= Why BRL-CAD =
 
= Why BRL-CAD =
 +
After participating in GsoC 2013 with BRL-CAD, I fell in love with CAD and will love to spend my summer contributing code to improve BRL-CAD software.
  
 
= Why Me =
 
= Why Me =
 +
First of all hailing from Africa with lack of computing infrastructure posed a lot of great challenges to  ascend to hackerdom especially with the scarcity of good programmers and lack of Internet access. Working with BRL-CAD in 2013 taught me a great deal especially working with the brightest researchers in this field. I believe my participation in GSoC this year would not only give me a great learning experience but also heighten my ambition of being a great Computer Science researcher in Africa. I see this  project as both an intellectual challenge to impact change through open source software.
 +
 +
= Contributions =
 +
 +
 +
== Past Coding Work ==
 +
 +
*GSoC 2013:
 +
**http://www.google-melange.com/gsoc/project/code_samples/google/gsoc2013/ch3ck/5721450489053184
 +
 +
*Xorg Evoc 2014
 +
**https://github.com/Ch3ck/xorg-xserverhttps://github.com/Ch3ck/xorg-xserver
 +
 +
*Other Projects
 +
**https://github.com/Ch3ck/SAMS
 +
**https://github.com/openmrs/openmrs-core
 +
 +
 +
==Current OpenSCAD work ==
 +
 +
Here are my current and ongoing contributions on OpenSCAD
 +
* Rework AbstractFunction::evaluate to virtual function only fixing calls
 +
**https://github.com/openscad/openscad/pull/1285

Latest revision as of 02:33, 6 April 2015

Project Title[edit]

Survey of CSG Algorithms


Personal Information[edit]

Name: Nyah Watad Check. Email: check.nyah@gmail.com. IRC Nick: Ch3ck, Ch3ck_.

Background Information[edit]

I am a third year Computer Engineering student of the University of Buea and one of the two first GSoC participants in Francophone Africa; worked with BRL-CAD in GSoC 2013, participated in the Google Doc Camp where I documented the BRL-CAD, OpenMRS and GNome projects. Participated as an X.org Evoc participant with X.org in 2014. I participated in as a Mentor in the 2014 Google Code-In for BRL-CAD.

Programming Background[edit]

  • X.org Evoc Participant, X.org, July – October 2014
    • Added Shatter Support to the X server project using Xephyr.
    • Clip Impedance layer to Xephyr and ran unit tests like Xts fixing bugs
  • Google Summer of Code Participant, BRL-CAD, April - September 2013
    • Implemented a pull routine to reverse the effects of a push on Geometry.
    • Integrated command into MGED interface. (C, XML, 500+ lines of code)
    • Tested the polynomial and matrix math routines, improving the speeds of the inverse matrix and matrix determinant routines by over 10%. (C, 300+ lines of code)
    • Integrated the xpush routine which pushes objects having more than a single matrix transformation into the push which pushes object matrix transformations to leaf nodes. ( C, 1000 lines of code.)
  • Book Sprint Participant, Google Doc Camp, October 2013
    • Participated in creating a contributors guide to BRL-CAD which will encourage a greater number of contributors to working on the BRL-CAD project.
  • SKILLS
    • Languages: C(Excellent), Java( Excellent ), C++( Beginner), Bash(Excellent), SQL(Proficient)
    • Tools: Secure Shell, subversion, Git, Linux, Netbeans, gdb, valgrind.

Synopsis/ Proejct Summary[edit]

This project aims at performing a comparison studies of 2D and 3D CSG algorithms, looking at the efficiency and robustness vis a vis the current implementations in OpenSCAD. Constructive Solid Geometry being one of the primary modeling techniques in the CAD industry is a major part of OpenSCAD. This involves studying the Clipper and CGAL library implementations in OpenSCAD and coming up with improvements on existing implementations.

Detailed Project Description[edit]

Introduction[edit]

Constructive Solid Geometry is a technique used in Solid modeling which allows a modeler to create a complex surface using Boolean operators to combine objects. This is often referred to as procedural modeling and can be performed on polygonal meshes. Here, the simplest solid objects used for representation are called primitives which are used to construct more complex objects using allowable Boolean operators such as union, intersection, difference and well as geometric transformations on those sets of objects.

OpenSCAD uses the Clipper Library to implement CSG in 2D and CGAL libraries to model 3D surfaces. My objective in this study is to look at the CSG Algorithms implemented by OpenSCAD and do a comparative study with other CSG algorithms, performing and algorithmic analysis and implement prototypes which can be useful in the OpenSCAD framework. This entails studying 4 main research papers together with various implementations of the algorithms associated with them and implementing the most efficient on the OpenSCAD platform.


CSG Algorithm for OpenSCAD(Current Implementation)[edit]

OpenSCAD is a parametric CSG based modelling software; which means objects are manipulated by editing the CSG descriptions of objects being rendered. It employs two different mechanisms for processing CSG descriptions. It uses the OpenCSG library for visualization during editing and the CGAL library for export created a meshed BREP output. However, the OpenCSG Library uses the Goldfeather and SCS algorithms to evaluate the 2D images of a CSG object from the given perspective. This is beneficial because of its speed and ability to render complex images within a short time.

However, CGAL employs the Nef Polyhedra for evaluating boolean operations on CSGs. It is robust and is able to handle non manifold input but proves incredibly slow in practice and not very suitable for time critical applications . There are also overflow issues that arise with using large size polyhedra. The used of BSP tree proves more efficient for CSG evaluations than CGAL's Nef based polyhedra . Below describes the use of BSP based algorithm for CSG evaluations than CGAL's Nef-based Polyhedra system.

BSP Tree based Implementation of Boolean Operations[edit]

Bernstein et all[1] presented an exact and more efficient method for evaluating 3D polyhedra already supported by OpenSCAD. Results from tests showed for a BSP tree implementation which 16-28X faster at performing iterative computations than CGAL's Nef polyhedra . The use of a BSP-tree based boolean algorithm allows the explicit handling of all geometric degeneracies without treating a large number of cases. There are also efficient CSG to BSP algorithms which could be used for easy representations in both formats.

BSP trees afford an alternative to B-rep algorithms that avoid their concomitant case explosion by explicitly handling all degenerate configurations of geometry. They have demonstrated to have suitable performance for interactive volumetric sculpting.

  • Numeric Substrates

In this system, a plane is a quadruple of floating point numbers interpreted as coefficients of a plane equation.

    • Concidence: Two planes p, q are coincident if and only if the determinant of all 2x2 minors are 0
    • coincident orientation: if p and q are coincident and Pa.Qa, Pb.Qb, Pc.Qc, Pd.Qd are non negative.
    • Point Validity: if a point A(p,q,r) is valid if i'ts determinant is non zero
 Pa Pb Pc
 Qa Qb Qc
 Ra  Rb Rc
    • Orientation: Given a point A(p, q, r) is valid and it lies behind, on, or in-front of the plane S if and only iff the following expression is negative, zero or positive
                            Pa Pb Pc  Pd
Pa Pb Pc                    Qa Qb Qc Qd
Qa Qb Qc       *            Ra Rb Rc Rd
Ra Rb Rc                    Sa Sb Sc  Sd

These predicates are implemented as static filtered floating-point predicates in the style of shewchuk, but deviations will be done to support double precision.

  • Geometric Substrates

In the preceding step, we defined planes as primitives, points as triples of planes and 4 predicates operating on these plans; now we define a convex polygon type for a constructor and splitting routine.

    • Convex polygon: This is a plane of support, s with a bounding plane {Bi} I £ Z. This vertices given by Vi = (s, Bi-1, Bi).
    • Construction of a Convex Polygon from a Plane: Given a plane h, this operation constructs a convex polygon representing h clipped by a very large box. The output polygon serves as a stand in for the infinite extent plane. A consistent axis aligned “very large box” is used for all calls to the constructor and consists of volume spaced bounded by planes X+, X-, Y+, y- , Z+, Z-.
    • Splitting a Convex Polygon by a Plane: Since polygon splitting may be viewed as two successive and complementarty instances of polygon clipping. The algorithm below similar to the Sutherland-Hodgman style polygon clippers, rather than deciding if and when to insert a crossing point to the output stream, we decide if and when to insert the splitting plane into the output stream.

Algorithm[edit]

If s is concident with h

 if s is similarly oriented to h
     Return (s, {Bi}i e Zn)
 Else 
    Return nothing

Else

  For I e Zn(in order)
     Output planes as specified by table lookup using:
      o(s, bi-2, bi-1, h)
      o(s, bi-1, bi, h);
      o(s, bi, bi+1, h)
   Return (s, output)
                 Algorithm: Clip(s, {bi}i e Zn) by h

The Above algorithm works in interoperability between input and output that is between the Point-based polygon soup and the Plane-based Representation working with Boolean operations.

Sample Algorithms/Links[edit]

  • Gilbert Bernstein and Don Fussell: Fast, Exact, Linear Booleans
  • Sebastian Steuer: Methods for Polygonalization of a Constructive Solid Geometry Description in Web-based Rendering Environments

Testing and Verification[edit]

Test would be written to verify and test the robustness of the BSP-tree based algorithm for CSG evaluations. This could apply Octoball tests from single operations between sizable meshes to Random Boxes tests and Sculpt Bunny depending on the efficiency. Also, the heat sink test which has proven to have a worst case performance of O(n2) in the size of input can be applied to the aforementioned implementation. Unit tests will be developed to handle different aspects of the BSP-tree evaluations and appropriate documentation attached for later works on the system. Also, any bugs coming up during this process will be fixed and documentation of this work done during development since OpenSCAD is actively developed and this will keep other developers informed on any changes taking place in the CSG module for OpenSCAD.

Deliverables[edit]

  • A paper containing the CSG Algorithm survey and it's relation to OpenSCAD
  • Prototype implementations integrated with OpenSCAD on the more efficient and robust CSG Algorithms.
  • An Evaluation of how more advanced modeling techniques could be realised with each algorithm.

Development Schedule/Timeline[edit]

This is a tentative plan which will be modified and developed as GSoC proceeds.

  • May 17 – June 06( 3 weeks)
    • Study papers on this topic
    • Discuss with developers on implementation specifics and further clarifications on CSG Algorithms supported by OpenSCAD.
  • May 24 – June 13 ( 3 weeks)
    • Check current implementation of the BSP tree algorithm for evaluating CSGs and porting to OpenSCAD.
    • Implement the BSP-tree representations in OpenSCAD or optimize current implementation.
    • Testing, debugging and documentation.
  • June 14 – June 20( 1 week)
    • Unit tests for BSP-tree routines
    • Testing degenerate cases.
  • June 21 – July 4 ( 2 Weeks) Mid Term Evaluations
    • Add functionality for CSG to BSP-tree conversion
    • Testing for robustness and degenerate cases.
  • July 5 – July 11( 1 week)
    • Implement testing suite for BSP tree implementation of the Bernstein algorithm.
    • Adding documentation to OpenSCAD framework
  • July 12 – July 25 ( 2 weeks)
    • Implement and test subroutines for Numeric substrates.
    • Testing and debugging.
    • July 26 – August 8 ( 2 Weeks)
    • Add functionality for Geometric Substrates together with sub routines.
    • Testing and debugging of Convex polygon constructor for clipped plane
  • August 9 – August 22( 2 Weeks)
    • Inside-out tests, address robustness issues with modules and improve performance of libraries.
    • Write paper on Comparison of CSG based algorithms in relation to OpenSCAD Implementations.
    • Check code for memory leaks and performance analysis, with unit tests added for implemented modules.
  • August 23 – August 29( 2 weeks)
    • Pencils Down, Code clean up.
    • Review documentation and Finalize paper writing.
    • Final evaluation, Submission of code to melange.

Time availability[edit]

I would be able to offer over 40 hours on the project. However, since the project would start during our second Semester; I would be coding mostly during the nights up to late June or early July when our semester ends. Also, to meet up with the demands of the project, I would be coding during weekends and regularly informing my mentors on the status of the project and regularly updating my logs in this respect.

Why BRL-CAD[edit]

After participating in GsoC 2013 with BRL-CAD, I fell in love with CAD and will love to spend my summer contributing code to improve BRL-CAD software.

Why Me[edit]

First of all hailing from Africa with lack of computing infrastructure posed a lot of great challenges to ascend to hackerdom especially with the scarcity of good programmers and lack of Internet access. Working with BRL-CAD in 2013 taught me a great deal especially working with the brightest researchers in this field. I believe my participation in GSoC this year would not only give me a great learning experience but also heighten my ambition of being a great Computer Science researcher in Africa. I see this project as both an intellectual challenge to impact change through open source software.

Contributions[edit]

Past Coding Work[edit]


Current OpenSCAD work[edit]

Here are my current and ongoing contributions on OpenSCAD