Computer Graphics - Homework Assignment 5 - Meshes

Overview:

In this assignment, you will be implementing a bit of mesh processing. You will tessellate a couple of shapes (cylinder and sphere), which means that you will create triangle meshes for them. You will compute the axis-aligned bounding box of a mesh and transform it to the unit cube. You will implement an algorithm for computing normals for meshes which don't have them. You will interact with a half-edge data structure. Your tessellation will create shapes like this:

Goals:

• Understand how to create and process triangle meshes.

• Understand how to tessellate surfaces, including position, normal, and texture coordinates.

• Understand how to compute the bounding box of a mesh.

• Increase familiarity with transformation matrices.

• Understand how to compute normals from a mesh with only surface positions.

• Understand how to work with a half-edge data structure.

• Become more comfortable with C++.

Background

• Book (FoCG,4e): Chapter 12.1 Triangle Meshes.
• Video: "Lecture 11: Geometry Processing" and "Assignment 5: Meshes"
• Quiz: Mesh Processing

(FoCG,4e is Fundamentals of Computer Graphics (4th edition) by Steve Marschner and Peter Shirley, 4th edition.)

Getting Started & Handing In:

• Download or clone this code repository. Don't fork it on GitHub, or else your code will be visible to everyone.

• Follow the instructions to install a working development environment: https://github.com/yig/graphics101 . You do not need to install Qt or any other external libraries for this assignment.

• The program is a command line program. The code for this project makes use of only the C++ standard library and glm, a library which provides the vector and matrix data types from the OpenGL Shading Language (GLSL).

• You are encouraged to write helper functions as you like.

• Build and run the code. The code should compile, but it will complain when running about not having enough arguments. You should see a message like:

    Usage: ./meshes cylinder #slices #stacks output.obj
    Usage: ./meshes sphere   #slices #stacks output.obj
    Usage: ./meshes cone     #slices #stacks output.obj
    Usage: ./meshes torus    #slices_big_circle #slices_small_circle tube_radius output.obj
    Usage: ./meshes cube     #slices_edge output.obj
    
    Usage: ./meshes normals          weights input.obj output.obj (weights must be either 'unweighted' or 'angle_weighted')
    Usage: ./meshes normals_halfedge weights input.obj output.obj (weights must be either 'unweighted' or 'angle_weighted')
    
    Usage: ./meshes normalize input.obj output.obj
    
    Usage: ./meshes transform M11 M21 M31 M41 M12 M22 M32 M42 M13 M23 M33 M43 M14 M24 M34 M44 input.obj output.obj (Mij are the elements of a 4x4 matrix)
  

• Add your code to mesh.cpp, shape.cpp, and halfedge.cpp. You may add helper functions if you wish.

• Build and run and test that it is working correctly. Qt Creator has a great debugger.

• Check your work with the command line tests by running ctest --verbose. This will test whether the following functions are working correctly:

  • computeNormals()
  • computeNormalsHalfEdge()
  • normalizingTransformation()
  • applyTransformation()
  • Cylinder::tessellate() (partial coverage; doesn't test whether triangles connect the right vertices)
  • Sphere::tessellate() (partial coverage; doesn't test whether triangles connect the right vertices)

• Run the following commands. Copy your output .obj files into a new output subdirectory named output. You can do this automatically by running the CMake target output via make output or cmake --build . --target output.

    ./meshes cylinder 5 10 cylinder-5-10.obj
    ./meshes cylinder 30 2 cylinder-30-2.obj
    ./meshes cylinder 3 2 cylinder-3-2.obj
    ./meshes sphere 5 30 sphere-5-30.obj
    ./meshes sphere 30 30 sphere-30-30.obj
    ./meshes sphere 3 1 sphere-3-1.obj
    ./meshes normals unweighted cube_binary.obj cube_binary-unweighted.obj
    ./meshes normals_halfedge unweighted cube_binary.obj cube_binary-unweighted-halfedge.obj
    ./meshes normals unweighted bunny.obj bunny-unweighted.obj
    ./meshes normals_halfedge unweighted bunny.obj bunny-unweighted-halfedge.obj
    ./meshes normals unweighted cylinder_irregular.obj cylinder_irregular-unweighted.obj
    ./meshes normals_halfedge unweighted cylinder_irregular.obj cylinder_irregular-unweighted-halfedge.obj
    ./meshes normalize bunny.obj bunny-normalize.obj
    ./meshes transform 0 0 1 0 0 1 0 0 -.1 0 0 0 0 0 0 1 cylinder-30-2.obj cylinder-30-2-transform-yzsquash.obj
  

  The following are bonus:

    ./meshes cone 5 10 cone-5-10.obj
    ./meshes cone 30 1 cone-30-1.obj
    ./meshes cone 3 1 cone-3-1.obj
    ./meshes torus 5 10 .25 torus-5-10-.25.obj
    ./meshes torus 20 3 .1 torus-20-3-.1.obj
    ./meshes torus 3 3 .25 torus-3-3-.25.obj
    ./meshes cube 2 cube-2.obj
    ./meshes cube 10 cube-10.obj
    ./meshes normals angle_weighted cube_binary.obj cube_binary-angle_weighted.obj
    ./meshes normals_halfedge angle_weighted cube_binary.obj cube_binary-angle_weighted-halfedge.obj
    ./meshes normals angle_weighted bunny.obj bunny-angle_weighted.obj
    ./meshes normals_halfedge angle_weighted bunny.obj bunny-angle_weighted-halfedge.obj
    ./meshes normals angle_weighted cylinder_irregular.obj cylinder_irregular-angle_weighted.obj
    ./meshes normals_halfedge angle_weighted cylinder_irregular.obj cylinder_irregular-angle_weighted-halfedge.obj
  

• You are encouraged to share blooper images you create while implementing the assignment on Piazza.

• Create a file named Notes.txt in the folder. Describe any known issues or extra features. Name people in the class who deserve a star for helping you (not by giving your their code!).

• When done, zip your entire meshes directory, your Notes.txt file, and your output directory as meshes.zip. There is a target named zip that will do this for you (cmake --build . --target zip) or you can use the cpack command from inside your build directory. The zip target will automatically run the output target, whereas you must manually run the output target if using cpack. If you create the zip file manually, do not include your build directory or the examples directory. They are large and unnecessary. Upload your solution to Blackboard before the deadline.

• THIS IS AN INDIVIDUAL, NOT A GROUP ASSIGNMENT. That means all code written for this assignment should be original! Although you are permitted to consult with each other while working on this assignment, code that is substantially the same will be considered cheating. In your Notes.txt, please note who deserves a star (who helped you with the assignment).

Rubric:

• (60 points) Tessellation. Create a triangle mesh for a shape, including positions, normals, and (optionally) texture coordinates. Your code goes in shape.cpp. The function signature is

    void Shape::tessellate( Mesh& mesh_out );
  

  For each type of shape, you will fill out the Mesh object passed into the function. Its properties are:

  • (20 points)
    mesh.positions: the array of 3D positions
    mesh.face_positions: the array of triangles indexing mesh.positions

  • (10 points)
    mesh.normals: the array of 3D normals
    mesh.face_normals: the array of triangles indexing mesh.normals

  • (bonus 10 points)
    mesh.texcoords: the array of 2D texture coordinates (aka explicit coordinates). Texture coordinates range from [0,1] in each dimension and are used to look up values in an image as a fraction of the image width and height.
    mesh.face_texcoords: the array of triangles indexing mesh.texcoords

  The shapes are:

  • Cylinder. The cylinder axis runs along the z-axis, from (0,0,0) to (0,0,1). The cylinder has radius 1. The cylinder has no top or bottom cap. The first texture coordinate runs around the circle of the cylinder and is equal to the angle as a fraction of 2π. The second texture coordinate runs up the cylinder. It is equal to z. There are two member variables available to the function, int m_slices and int m_stacks. Your cylinder should have m_slices vertices around the circle (so three is the minimum) and m_stacks vertices along the axis (so two is the minimum). There will be a total of m_slices*m_stacks positions, m_slices normals (since each stack has the same normals), and (m_slices+1)*m_stacks texture coordinates (since the last texture coordinate of each circle must be 2π/2π = 1 and so can't be 0 like the first or else the texture would tear).

  • Sphere. The sphere is the unit circle centered at the origin (0,0,0) with radius 1. The first texture coordinate should be the longitude angle as a fraction of 2π. The second texture coordinate should be the latitude angle as a fraction of π. There are two member variables available to the function, int m_slices and int m_stacks. Your sphere should have m_slices lines of longitude (so three is the minimum) and m_stacks lines of latitude (not counting the north and south poles, so 1 is the minimum). There will be a total of m_slices*m_stacks + 2 positions (the +2 comes from the north and south pole), the same number of normals as positions, and (m_slices+1)*m_stacks + 2 texture coordinates (for the same reason as the cylinder, +2 for the north and south poles).

  • (bonus, quarter credit) Cone. The cone whose axis runs from (0,0,0) to (0,0,1) with radius 1. The cone has no bottom cap. The first texture coordinates runs around the circle of the cone. The second texture coordinate runs up the cone. It is equal to z. There are two member variables available to the function, int m_slices and int m_stacks. Your cone should have m_slices vertices around the circle (three is the minimum) and m_stacks vertices along the axis (not counting the top point, so 1 is the minimum). There will be a total of m_slices*m_stacks + 1 positions (the +1 is the top point), m_slices normals (since each stack has the same normals), and (m_slices+1)*m_stacks + 1 texture coordinates (for the same reason as the cylinder and sphere, +1 for the top point).

  • (bonus, quarter credit) Torus. A torus is a donut. For our "unit" torus, the center of the doughy tube is the unit circle in the xy plane with radius 1. If you cut into the tube of the donut, you will cut through a circular cross section. The cross section has radius m_cross_radius. You can think of it as if dough with constant thickness m_cross_radius was added to the unit circle in the xy plane. The first texture coordinate runs around each cross section circle. The second texture coordinate runs around the tube. There are three member variables. real m_cross_radius was already mentioned. The other two are int m_slices_tube_circle and int m_slices_cross_circle. Your torus should have m_slices_cross_circle vertices around each cross-section of the tube, and m_slices_tube_circle vertices around the entire tube. (Three is the minimum for each.) There will be a total of m_slices_tube_circle*m_slices_cross_circle positions and normals, and (m_slices+1)*(m_stacks+1) texture coordinates (since the last texture coordinate of each circle, tube and cross section, must be 2π/2π = 1 and so can't be 0 like the first or else the texture would tear).

  • (bonus, quarter credit) Cube. A unit cube whose sides are axis-aligned and whose corners are (-1,-1,-1) and (1,1,1). Each square face should have the entire texture mapped onto it. There is a single member variable, int m_slices_edge. Each edge of your cube should have m_slices_edge vertices (so two is the minimum), and each face of your cube should have m_slices_edge*m_slices_edge vertices. There will be a total of (m_slices_edge-2)*(m_slices_edge-2)*6 + (m_slices_edge-2)*12 + 8 positions (since each edge is shared by two faces, and each corner is shared by three faces). There will be a total of 6 normals (one per face). There will be a total of m_slices_edge*m_slices_edge texture coordinates.

• (30 points total) Mesh normals. Your code goes in mesh.cpp and (a tiny bit) halfedge.cpp. Given a Mesh object with positions but no normals, compute a normal for each position by averaging the (flat) normal of each triangle touching the vertex. You can do this by initializing a running tally of the average normal for each vertex. Each triangle uses its positions to compute a normal via the cross product. Add this normal to the running tally for the vertex. If the strategy parameter is Unweighted, simply normalize the normal before adding it to the running tally. For bonus, if the strategy parameter is AngleWeighted, normalize the normal and then scale it by the angle of the triangle at the vertex before adding it to the running tally.

  • (10 points) Without a half-edge data structure. Iterate over each triangle. Add its unweighted normal to the running tally for each of its three vertices. The function signature is:

      void Mesh::computeNormals( MeshNormalStrategy strategy )
    

    • (bonus 5 points) Add each triangle's AngleWeighted normal to the running tally.

  • (20 points) With a half-edge data structure. Iterate over each vertex. Call halfedges.vertex_face_neighbors( vertex_index ) to obtain the triangles touching a vertex as a vector of indices into .face_positions. Then iterate over the triangles and add their unweighted normals to the running tally for the vertex. The function signature is:

      void Mesh::computeNormalsHalfEdge( MeshNormalStrategy strategy )
    

    Note that the half-edge function .vertex_face_neighbors() is not written for you. You must fill in the code for this function in halfedge.cpp. It is a very short function. Follow the provided outline. Consult HalfEdgeTriMesh::vertex_vertex_neighbors() for reference; it differs from .vertex_face_neighbors() by exactly one line.

    • (bonus 5 points) Add each triangle's AngleWeighted normal to the running tally.

• (10 points) Mesh normalization. Your code goes in mesh.cpp.

  • Compute the 4x4 transformation matrix that, if applied to the .positions of the mesh, translates and uniformly scales it to tightly fit within the unit cube. The unit cube is the cube with axis-aligned faces centered at the origin with side length 2, aka corners at (-1,-1,-1) and (1,1,1). The function signature is:

      mat4 Mesh::normalizingTransformation() const
    

    normalizingTransformation() does not affect any vertex positions. (That is applyTransformation(), described below.) You can break down the job of normalizingTransformation() into two parts:

    1. Compute the axis-aligned bounding box for all positions in the mesh. To do this, find the minimum and maximum x, the minimum and maximum y, and the minimum and maximum z among all positions in the mesh. The minimum x,y,z is one corner of the bounding box and the maximum x,y,z is the other corner.

    2. Compute the transformation that translates this bounding box to be centered at the origin. (Hint: The center of the bounding box is 0.5*( min_xyz + max_xyz ).) Compute the transformation that uniformly scales the bounding box to fit inside the [-1,1] cube, which has edge length 2. (Hint: max_xyz - min_xyz is the size of the bounding box in all three dimensions. Scale so that the maximum size in any dimension equals 2.) If your translation matrix is T and your scale matrix is S and a vertex is v, the transformation you want to apply is S * T * v, which means that the box will first be centered at the origin and then scaled so that its maximum dimension exactly fits within [-1,1].

  • You must also write a function which applies a transformation matrix to the mesh. You apply the transformation matrix itself to .positions. Don't forget that you apply the inverse transpose of the upper-left 3x3 portion of the matrix to the normals. The function signature is:

      void Mesh::applyTransformation( const mat4& transform )
    

Tips

• There is some relevant pseudocode in the docs directory. Consider that a cylinder, sphere, cone, and torus are all rectangles rolled up in different ways, with one edge glued (two in the case of the torus) and possible a north or south pole attached.

  • The make_mesh_grid() pseudocode is an example showing you how to create a square out of triangles, where the number of triangles can be specified by the user. The square is a 2D square in the x,y plane, so all vertices have z=0. This pseudocode should help you understand how to create a mesh, but it is a simpler scenario that what you have to do in this project in two major ways.
    • One difference is that you can make a square without defining a separate explicit coordinate system for the shape, like latitude and longitude for spherical coordinates or theta and z for cylindrical coordinates. That's because x,y coordinates parameterize (or walk along) both the square and the output positions simultaneously. In the general case, you iterate over the explicit coordinates (e.g. spherical or cylindrical) and use a formula to convert from, say, latitude and longitude to x,y,z coordinates. Slices and stacks tell you how many times to iterate along the two explicit dimensions (e.g. latitude and longitude). For a cylinder, you can think of this as rolling the square into a tube.
    • The second difference is that a sphere and a cylinder are watertight shapes. You will need to create faces between the last and first slice. For the sphere, you will also need to create an extra vertex at the top and bottom (the north and south poles) and then faces to glue the top and bottom stacks to those vertices.

You need to fill our mesh.normals when tessellating these shapes. The computeNormals() function is for arbitrary triangle meshes you load which don't already have normals. When tessellating, you know the shape and so you have a formula for the correct normal at each position. This is better than computing the normal at each face or by averaging face normals.

• For debugging an OBJ that you program saves, I highly recommend the free program meshLab (http://www.meshlab.net/):

  

  • A visualization like the above tells you a lot about your mesh. With a few viewer settings, you can see the UV coordinates. You can see the x,y,z axes and tell that the mesh is running along +z. You can see the normals (face and vertex). You can see that the faces have counter-clockwise orientation because "back-faces culling" is on and the back-faces are missing. To create a similar visualization, open your OBJ file with meshLab and then, under the Render menu, choose:

    • Show Normal/Curvature
    • Show Axis
    • Show UV Tex Param
    • Render Mode > BackFace Culling

  • To load a texture, like uv_grid.png or earth.png, go to Filters > Texture > Set Texture.

  • To debug normals, it can be helpful to view with mesh with "reflection lines", which renders the mesh as if it were a mirror in a long, striped corridor. In meshLab, choose "Render > Render Mode > Smooth" and choose "Render > Shaders > reflexion_lines".

• A cylindrical coordinate system is (r,φ,z), where r is the radius, φ is the angle around the circle of the cylinder, and z is the cartesian z coordinate. For your cylinder, r = 1, slices evenly sample φ, and stacks evenly sample z.

  • To convert cylindrical coordinates to Cartesian coordinates for your positions:
    x = r cos( φ ), y = r sin( φ ), and z = z.

  • A cylinder's normal at a position has the same x,y as the position, but z = 0.

  • A cylinder's texture coordinates are: ( φ/(2π), z ).

• A spherical coordinate system is (r,θ,φ), where r is the radius, θ ∈ [0,π] is the angle away from the north pole (like latitude), and φ ∈ [0,2π] is the angle around the sphere like longitude. For your sphere, r = 1, slices evenly sample φ, and stacks evenly sample θ.

  • To convert spherical coordinates to Cartesian coordinates for your positions: x = r sin(θ) cos(φ), y = r sin(θ) sin(φ), and z = r cos(θ).

  • A sphere's normal at a position has the same x,y,z as the position.

  • A sphere's texture coordinates are ( φ/(2π), 1 - θ/π ).

• For a torus, a convenient coordinate system is 2D polar coordinates (r,θ) along with a cartesian z coordinate. In this way, the formula for a torus given u,v parameters is (r,θ,z) = ( 1 + p cos( 2 π u ), 2πv, - p sin( 2 π u ) ), where p = m_cross_radius. The u,v parameters go from 0 to 1 around the cross-section and tube, respectively. You can convert the 2D polar coordinates to 2D cartesian coordinates as x = r cos(θ) and y = r sin(θ). The normal is the position after translating the center of the cross section to the origin, which means subtracting 1 from r. To get a unit normal, drop the p scale factor. (You can verify that it is unit by making use of the identity cos²(θ) + sin²(θ) = 1.)

• When writing Mesh::normalizingTransformation(), you can use glm::translate(mat4(1.0), vec3(x,y,z)) to get a translation matrix that translates by x,y,z. You can use glm::scale(mat4(1.0), vec3(s,s,s)) to get a scale matrix that scales uniformly by s. You can also build these matrices yourself quite easily:

  • A uniform scale matrix is a matrix whose diagonal is the value to scale by. You can create a 3x3 uniform scale matrix via mat3( scale_factor ), where scale_factor is a number. You can create a mat4 from a mat3 m via the constructor mat4(m). You can create an identity mat4 via mat4(1.0). You can create a translation matrix by setting the fourth column of an identity mat4 t via t[3] = vec4( x,y,z,1.0 ) or, equivalently, t[3] = vec4( p,1.0 ), where p is a vec3.

• The glm math library has many useful functions. It implements the vector and matrix data types from the OpenGL Shading Language (GLSL). In other projects, you've used it to normalize() a vector and compute the dot() product between two vectors. You can also use it to compute the cross() product of two vectors.

• glm::min() and glm::max() operate component-wise. That means that glm::min( vec2( 5.0, -3.0, vec2( 1.0, 0.0 ) ) is vec2( 1.0, -3.0 ). This is useful for computing the axis-aligned bounding box of a set of points, which is what you need to do in Mesh::normalizingTransformation().

• When computing angle weighted normals, you need to compute the angle at each corner of the triangle. Recall that the dot product of two vectors a · b = cos(θ) ||a|| ||b||, where θ is the angle between them. So you can obtain the angle between two vectors via the inverse cosine (arccosine) of the normalized vectors. The function that computes arccosine is acos(). Call it on the dot product of two of the triangle's edges, normalized. Note that acos() is very strict about its input being between -1 and 1. Due to floating point issues, sometimes the dot product will produce a number epsilon out of range. Use glm::clamp() or min() and max() to ensure the value you pass to acos() is between -1 and 1.

• You can access the constant pi aka π = 3.14... via the constant pi (defined in the #include "types.h" header) or by calling glm::pi().