Assignment 4: Using the Stencil Buffer & GPU

The goal of this assignment is to become familiar with both classic graphics hardware (the stencil buffer) and the modern GPU (using programmable geometry & pixel shaders).

Tasks

• First, download and compile the provided files. The basic program will load an .obj into a scene. An appropriately sized quad that represents the floor is placed beneath the object and an appropriately sized quad that represent the mirror is placed to the left of the object. A yellow point above the object represents the position of a point light source. Pressing 'a' will toggle animation of this light, moving it in a circle above the object. Pressing the space bar will move the light one step along this path.

    ./render -input bunny_1k.obj
  

• Press 'm' to toggle stencil buffer mirror rendering, which is incomplete. You need to write code to "duplicate" and render the geometry of the mirrored object and floor as they appear to exist, reflected in the mirror. Pressing 'r' will toggle a visualization of the reflected geometries, rendered in blue. If everything is working with your stencil buffer, the mirror visualization should then work as shown in the picture below. The objects in the mirror are tinted slightly blue for effect.

  

  If you study the resulting imagery closely, you'll notice that the lighting is decidedly incorrect. The bunny in the mirror has different local shading than the physical bunny. Which light source should that bunny be lit by? For extra credit, describe the problem in your README.txt file and improve the local shading on the mirrored bunny. (Actually, the lighting on the physical bunny is wrong too and can be improved!)

• Next, let's implement shadow volumes. The first step is to determine the silhouette edges in the model. These are edges that have one neighboring triangle facing toward the light source, the other triangle facing away from the light source. Pressing 'e' will toggle the visualization of these edges in red, as shown below. You need to write the code to determine which edges are silhouettes and render this geometry.

  Once these edges are found, we can construct the polygons that form the "sides" of the shadow volumes. These polygons are extrusions of the silhouette edges away from the light source, towards infinity. Pressing 'p' will toggle the visualization of these polygons in green, as shown below, using blending so you can partially see through these polygons and see the depth complexity of the multiple layers of shadow volume polygons. You need to write the code to construct and render these polygons.

  

• The last step is to do the multi-step rendering to create the shadows. This involves juggling the frame buffer, depth buffer, and stencil buffer. The basic idea is to first render the scene without lighting. Then, we do a special rendering pass of the shadow volume polygons: everywhere the depth buffer passes, we will increment (if the shadow volume polygon faces toward the camera) or decrement (if the shadow volume polygon faces away from the camera) the stencil buffer. Finally, we render the scene a second time with the lights on, but only where the stencil buffer is zero. The shadow rendering mode is toggled by pressing 's'.

  An excellent stencil buffer tutorial written by Mark Kilgard at NVIDIA: http://developer.nvidia.com/object/Stencil_Buffer_Tutorial.html (alternate link) gives more detail on each step of the method.

  

• There are several extensions to the basic shadow volumes code that you can do for extra credit, including: handle non-closed objects, improve the efficiency of the method, make the shadow volumes and mirror rendering methods work together (the same shadows should appear in the physical scene and the mirror), or implement "Z-fail" shadow volumes so that the rendering is correct when the camera is inside of a shadow volume (create a new scene to demonstrate this extension). Also, for extra credit you can mirror the light source. Both the real object and the reflected object should be lit by two light sources, the real and the reflected light. Describe any extensions you make in your README.txt file.

• Next... let's experiment with GLSL shaders. Try the examples below, to verify that your system can render with shaders. Once loaded, press the 'x' key to initialize the shaders. Pressing 'x' again will disable the shaders.

  ./render -i cube.obj -v checkerboard.vs -f checkerboard.fs
  ./render -i sphere.obj -v orange.vs -f orange.fs 
  

  If your system supports GLSL and the noise function, the models should render like the images below.

  

  If the checkerboard looks good, but the orange is missing the bump map (if it just looks like a smooth orange ball), you card doesn't support the noise function (but you can still do some interesting things with shaders for the homework). Be sure to press 'n' to render with per-vertex smoothed normals.

  If you get it working quickly... great! However, if after a reasonable amount of effort (use Google, etc.) you are unable to get a working setup either on your own machine or a borrowed machine (try the VCC), then write up what you tried, and what errors you were seeing in your README.txt file. You can submit carefully written and commented but untested code for the remaining portion of this homework assignment and will receive full credit for your effort.

• First, poke around at the vertex & fragment shaders for these two examples. Make small modifications to the code and see what happens. You don't have to quit & restart the program when a shader changes, you can just press 'l' to recompile, relink, and reinitialize the shaders. Note: if you make a syntax error in the GLSL code, the program will most likely crash without giving you an informative error message. Thus, it's recommend to frequently test your code after writing each line or two, so you know exactly where the error was introduced.

• Once you're comfortable with the provided examples, start your own vertex & fragment shader to simulate wood grain. To start, mimic the rings of tree growth along a primary axis with alternating cylindrical bands of color. Then, you can use a noise function to add geometric variations to the bands to look more natural (so they aren't perfect circles). Experiment with subtle variations in the color as well. Finally, you can add variations in the normal to represent the ridges in the grain visible in the specular highlight of polished wood. In addition to the code, submit screenshots of your wood grain shader. Include references to any online resources you used for this portion of the homework.

• For extra credit, create another complex shader and submit that code and screenshots of the results. In your README.txt file describe your target result and what was necessary to create the result.

  Yet another option for extra credit is to mix the shaders with the stencil mirror and/or stencil shadows.

Provided Files (hw4_files.zip)

• Basic Code (CMakeLists.txt, argparser.h, boundingbox.cpp, boundingbox.h, camera.cpp, camera.h, glCanvas.cpp, glCanvas.h, main.cpp, matrix.cpp, matrix.h, and vectors.h)

  Similar to the previous assignments.

• Half-Edge Quad Mesh Data Structure (edge.cpp, edge.h, hash.h, mesh.cpp, mesh.h, triangle.h, and vertex.h)

  Similar to the triangle half-edge data structure you implemented in assignment 1.

• Rendering (render.cpp)

  The code for rendering the mesh with stencil buffer tricks.

• Shaders (load_shaders.cpp, orange.fs, orange.vs, checkerboard.fs, and checkerboard.vs)

  The code for loading, linking, & initializing GLSL shaders, and two example shaders.

• Test meshes (bunny_1k.obj, bunny_200.obj, bunny_40k.obj, cube.obj, and sphere.obj)