Advanced Computer Graphics Shadow Techniques

1
Advanced Computer GraphicsShadow Techniques

• CO2409 Computer Graphics
• Week 18

2
Lecture Contents

• Basic Shadows
• Pre-calculated Shadows
• Shadow Mapping
• Stencil Shadows

3
Basic Shadows

• Shadows in a scene to help resolve the relative
  positions of models
• A couple of basic techniques have long been used
• Draw a blob under a model
• Draw the model flattened (scaled to 0 in Y) on
  the floor
• Can project straight down or away from the light
• Both methods require a flat floor to work
  correctly

Positions Ambiguous?
Shadows Resolve
4
Basic Shadows

• These methods are still frequently used, why?
• Basic methods enough to resolve model positions
• Very cheap techniques good when many models
• More advanced techniques are complex and slow
• Advanced techniques often draw attention to
  themselves
• Sharp edges / too stark, problems in complex
  cases
• In open areas, shadows tend to blur towards blobs
  in any case
• Viewer may not notice / appreciate better accuracy

5
Pre-calculated Shadows

• However, improved shadows can give a better sense
  of space and/or atmosphere
• When used appropriately
• If we have static models and lights, we can
  pre-calculate the lighting in the scene
• A standard technique has been to use static
  shadow maps

• Darkening textures applied over the main model
  textures
• The maps are pre-calculated
• Often using radiosity methods
• Allows for subtle lighting effects

6
Pre-calculated Shadows

• More recent techniques perform an offline
  mathematical simulation of light transfer in a
  scene
• Reflections, refractions and transmission
• One example of this technique is Pre-computed
  Radiance Transfer (PRT)
• This results in a set of equations to calculate
  both lighting shadows to be computed in
  real-time

• Allows for dynamic lights
• And often some measure of model movement
• No model/model interaction

7
Dynamic Shadow Mapping

• Dynamic Shadow Mapping is an extension of
  render-to texture techniques used for shadows
• Also called perspective shadow mapping (PSM) or
  just shadow mapping
• The scene is rendered into a texture (a shadow
  map), but from the point of view of the light

• Then the scene is rendered normally, but each
  pixel first tested against shadow map
• The pixel is not lit if it is in shadow from the
  light

8
Shadow Mapping Method 1

• Create a render target texture for each light
• Each pixel in the texture is a single float
  value
• Not four R,G,B A values
• This is a floating-point texture. It will be the
  shadow map for the light

• Render the scene from the lights point of view
• Need to treat the light like a camera
• Result is effectively a depth buffer for the
  scene

9
Shadow Mapping Method 2

• Now render the scene normally
• But with an extra step to find shadowed areas
• After transforming each pixel, check its
  visibility from each light before lighting it
• Find depth of the pixel as seen from light
• Also transform the pixel into the lights
  viewport space to get depth stored in shadow map
• If the depth in the shadow map is less than the
  depth of the pixel, then something is obscuring
  the pixel and it must be in shadow
• Otherwise light it as normal

10
Shadow Mapping Diagram
11
Shadow Mapping Detail 1

• To treat a light as a camera, we need a view and
  projection matrix for it
• A spotlight is the simplest case with a position
  and a field of view (FOV) already

• A point light shines in all directions
• We render six images in a cube around the light
• FOV of 90 for each one
• An example of a cube map

12
Shadow Mapping Detail 2

• A directional light poses two problems
• No position as a camera. No FOV to use
• Solve first problem by positioning the
  directional light very far away outside the
  scene

• Solve the second by using an orthogonal
  projection matrix
• Project vertices from 3D to 2D along parallel
  lines
• Not towards camera point like standard
  perspective projection
• Dont need FOV

13
Shadow Mapping Issues

• Shadow mapping has two key problems
• Texels of shadow map can be clearly visible
• Polygons sometimes self-shadow
• Increase shadow map resolution for better quality
• But lower performance, increased memory
• Better, blur or soften the shadows / map
• Many methods here
• Self-shadowing resolved by tweaking calculated
  and compared depth
• Other solutions illustrated in lab

14
Stencil Shadows

• Another method for real-time shadows is Stencil
  Shadows, an advanced stencil buffer technique
• Stencil shadows gained great interest because of
  their much-hyped use in the game Doom3 (2003)

• But they have a longer history
• Shadow volumes first discussed by Frank Crow
  (1977)
• Stencil buffer use proposed by Tim Heidmann
  (1991)
• Used extensively Severance Blade of Darkness
  (1999)

15
Stencil Shadows Process 1

• Render scene with ambient light only
• For each light
• Find silhouette of each model from the light
• Extrude silhouette to make shadow volume
• Render shadow volume using stencil buffer only
• Result is non-zero stencil values where shadow
  intersects with scene (explained later)

16
Stencil Shadows Process 2

• For each light (continued)
• Render entire scene using only this light but
  only render if stencil is zero (not in shadow)
• Add the result colour to the existing colour
• Each lights contribution will be accumulated
  onto the basic ambient light to give final result

17
Stencil Operation Depth Pass

• Shadow volumes only drawn in the stencil buffer
• No colour or depth output
• But the depth buffer is tested
• Render in two passes
• Pass 1 Render back faces, increasing the stencil
  buffer values for each visible pixel
• Pass 2 Render front faces, decreasing the
  stencil buffer values for each visible pixel
• This is the depth-pass method (stencil updated
  where pixels pass the depth test i.e. visible)
• Diagrams follow

18
Depth-Pass - Diagrams

• Depth-Pass Shadow Volumes
• Front Faces 1
• Back Faces -1
• Result ! 0 in shadow

Problem with Depth-Pass Camera in Shadow Volume
19
Depth-Fail

• Depth pass has problems instead use depth-fail
• Pass 1 Render back faces, increasing the stencil
  buffer values for each non-visible pixel

• Pass 2 Render front faces, decreasing the
  stencil buffer values for each non-visible pixel
• Called Carmacks Reverse due to John Carmacks
  use in Doom3
• Actually, idea from video card companies (NVidia
  / Creative Labs)

20
Finding Silhouettes

• To create the shadow volume we need to find the
  silhouette of the model wrt the light
• The edges to extrude
• Can do this at run-time by comparing normals of
  adjacent faces with the light vector
• If one points away and the other towards then we
  have a silhouette edge
• Then dynamically create a shadow volume from the
  found edges
• This can be expensive
• CPU time dynamic resource creation

21
Pre-calculated Shadow Models

• Instead we can create a special shadow model
  associated with each model
• Copy the base model, but make the vertices in
  each face unique and set all vertex normals
  face normals
• Replace every edge in the model with a infinitely
  thin quad (pair of triangles)
• Doesnt change the model shape

Inserted quads shown stretched out for clarity
actually infinitely thin
22
Pre-calculated Shadow Models

• Use a vertex shader to extrude the shadow model
  vertices that point away from the light
• They will stretch out the quad-edges behind them,
  forming the required shadow volume