Assignment 1: Simplification and Subdivision Surfaces

The goal of this assignment is to try your hand at various mesh processing tasks. A half-edge mesh adjacency data structure is provided, along with a very simple .obj mesh parser. As you modify the triangle mesh, it is critical that the manifold properties of the input surface are maintained, and that the adjacency data structure remains consistent. Throughout this assignment you are asked to consider the efficiency of various mesh operations: is it quadratic, linear, logarithmic, constant, etc? Since this assignment is rather long, it's ok if you don't always implement the most efficient strategy (which may require additional data structures), but please discuss the tradeoffs in your README.txt file. Please use the provided README.txt template.

Warning: This assignment has one small component (Gouraud shading) & two large components (simplification and subdivision). Start early, and just get the basics working first. Spend a reasonable amount of time coding and write about what you've learned in your README.txt file. Also, to streamline grading, please indicate which portions of the assignment are finished & bug free (full credit), attempted (part credit) or not started (no credit) by filling in the provided hw1_gradesheet.txt and submitting it with your assignment. The instructor will then check, edit as needed, and finalize your assessment.

Tasks

• Download the provided source code and compile it as before. You should be able to load simple .obj files and turn wireframe visualization on & off by pressing "w". Try these command lines:

  ./mesher -input cube.obj -size 300 
  ./mesher -input cube.obj -size 300 -wireframe
  ./mesher -input open_box.obj -size 300 
  ./mesher -input open_box.obj -size 300 -wireframe
  

  Notice that the adjacency data structure detects boundaries in the mesh and draws them with thicker red lines in the wireframe visualization:

  

• Now you're ready to start coding. First make use of the adjacency data structure provided to implement gouraud shading (we'll talk about this in Lecture 3). In gouraud shading, average normals are computed for each vertex, these normals are used for lighting, and the shaded values automatically interpolated in GL. You can change the color and/or normal between each specified vertex. This shading option should be enabled by specifying -gouraud on the command line, or pressing "g".

  ./mesher -input bunny_1k.obj -size 500
  ./mesher -input bunny_1k.obj -size 500 -gouraud
  

  

  Note that the average normals for each vertex can be efficiently computed without any adjacency information at all, but that wouldn't give you practice with the data structure! For extra credit, don't average the normals across "sharp crease" edges (whose faces differ in normal by more than a user-defined threshhold angle).

  Comment on the efficiency & order notation of your implementation in your README.txt.

• Next, let's experiment with edge collapses, the key mesh manipulation operation from the Progressive Mesh framework. When you press the "d" key (for decimate, a.k.a. simplification) the number of triangles in your mesh should be reduced by approximately 10 percent.

  First implement the topology of an edge collapse. Identify a target edge to collapse. For now, just pick a random edge (or any edge) in the mesh. Using the adjacency data structure, identify which two triangles will be removed permanently from the model, and which other triangles will change. Choose a simple location for the remaining vertex (e.g., it's old position, or a position averaged with the deleted vertex). Note: in the provided adjacency data structure it is recommended that you will delete and then re-create triangles rather than modify them and the related edges. Initially you may want to just collapse one edge each time the "d" key is pressed.

  ./mesher -input bunny_1k.obj -size 500 -wireframe
  

  

  Even though a single random collapse works great, you will probably quickly run into problems: 1) the mesh looks bad with small triangles and big triangles that don't represent the high-resolution model very well, 2) it creates self intersections of the surface (it turns itself inside out and you can see the blue side), and 3) (worst-of-all) it sometimes crashes with an error like:

    assertion "opposite == NULL" failed: file "edge.h", line XXX
  

  Think about what that error might indicate is wrong with your implementation. (If you get a different error, of course, try to explain that one.) Certain edges should not be collapsed because it would cause the surface to become non-manifold. For example, the "ring" of vertices surrounding the target edge should be unique. In the diagram below, if vertices 1 and 4 refer to the same vertex, then after the collapse of the solid black edge, four triangles will meet at the same edge. This proposed edge collapse will alter the topology of the model and yield a non-manifold condition. Write code to check for these conditions. NOTE: This code can be tricky. It's ok if you only just catch some of these conditions and your program still crashes occasionally. Don't spend too long trying to make it perfect. Discuss this in your README.txt file.

  

  Now implement something to make a smarter choice for which edge to collapse (this might also minimize the occurance of lingering bugs from the previous part). Collapsing the shortest edge first (or actually make that the shortest legal edge) will usually do a very good job at preserving overall surface shape. Done naively, repeatedly selecting the shortest edge for collapse can be an expensive operation. That's fine for now, you can implement something very inefficient. Discuss these performance issues in your README.txt file and what common data structures would help.

  

• Finally, implement the Loop Subdivision Surface scheme (we'll talk about this in Lecture 4). When the "s" key is pressed, your implementation should perform one iteration of subdivision. First you want to get the topology correct. Add a vertex at the center of each edge, then replace each triangle with 4 smaller triangles. Make sure you share the new vertices across the edges and don't create multiple copies of the vertex. Consider how to do this efficiently and discuss various options in your README.txt file. The provided hash table structure for tracking the "parents" of each newly added vertex should be very helpful for this task. You can verify correct vertex sharing by visualizing the wireframe and looking for the red boundary edges. If you have extra red lines on your mesh (not at the boundaries) you have not properly shared the vertices.

  ./mesher -input open_box.obj -size 300 -wireframe
  

  

  Once you have the correct topology and vertex sharing, you can adjust vertex positions according to the Loop Subdivision rules. To handle general shapes you must implement the rules for regular vertices (valence = 6, i.e., 6 edges meeting at a vertex), extraordinary vertices (valence != 6), and boundary conditions. Here are the Loop Subdivision Rules from the SIGGRAPH 2000 course notes - Subdivision for Modeling and Animation (page 70).

  

  Finish off your subdivision surface code by implementing the simple binary (infinitely sharp or infinitely smooth) crease method described in: "Piecewise Smooth Surface Reconstruction." H. Hoppe, T. DeRose, T. Duchamp, M. Halstead, H. Jin, J. McDonald, J. Schweitzer, W. Stuetzle. ACM SIGGRAPH 1994.

  ./mesher -input creased_cube.obj -size 300 -wireframe
  

  

Ideas for Extra Credit

• Implement gouraud shading with a proper average weighted by angle for the triangles meeting at each vertex (if you haven't done this already) and a crease angle threshhold. Find or create a mesh which illustrates these differences.

• Implement a "better" edge weight selection method. First define your goal for simplification and construct an edge weight function that matches that goal. E.g., if your goal is to create equilateral triangles than the shortest edge collapse is a very good choice. Discuss in your README.txt.

• Implement a better positioning method for the remaining vertex in an edge collapse. Discuss in your README.txt.

• Implement the geomorph option to smoothly transition between two different resolutions within a progressive mesh.

• Improve the efficiency of your implementation by using the right data structures for various operations (hash tables, priority queues, etc.)

• Implement the integer crease weight control described by "Subdivision Surfaces in Character Animation", DeRose, Kass & Truong, SIGGRAPH 1998 to perform n iterations of crease subdivision rules followed by infinite iterations of smooth subdivision.

• Implement the floating point weight control by interpolating between the nearest integer iterations.

• Implement the butterfly interpolating subdivision surface scheme.

Provided Files (hw1_files.zip)

• Basic Code (vectors.h, matrix.h, matrix.cpp, argparser.h, camera.h, camera.cpp, glCanvas.h, glCanvas.cpp, main.cpp, Makefile, MersenneTwister.h)
  Similar to Homework 0.

• Mesh Library (boundingbox.h, boundingbox.cpp, vertex.h, triangle.h, edge.h, edge.cpp, mesh.h, mesh.cpp)
  An implementation of the half-edge data structure. The load function will parse simple .obj files.

• hash.h
  For efficient, constant time access of parts of the data structure, we store edges and vertex ancestry in a hashtable. It is used by the Mesh code to efficiently find pairings between opposite edges at load time. Unfortunately, the hash map or unordered map is not quite standard in STL, and depending on your platform, you may need to tweak this file.

• .obj files (cube.obj, creased_cube.obj, open_box.obj, bunny_200.obj, bunny_1k.obj, bunny_40k.obj, cactus.obj, knot.obj, and hypersheet.obj)
  Some models from: "Piecewise Smooth Surface Reconstruction.", Hoppe et al, 1994
  

Please read the Homework information page again before submitting.