 |
» |
|
|
|
This section contains more detailed discussions of some selected
texture mapping-related topics below: Discussion: MIP Map | |
"MIP" stands for multum in parvo; literally,
"many things in a small place." Why use MIP maps? MIP maps are generally used to reduce the aliasing effects
that naturally take place with texture mapping. Because the cost
and repetition of anti-aliasing calculations are high, pre-computing
their values in the form of a MIP map can dramatically reduce the
overall performance cost of anti-aliasing. When a texture map is accessed for a single screen pixel,
one of three things can happen: The pixel maps to less than one texel
(this means that several screen pixels map to a single texel of
the texture map.) The texel information must be "magnified" (by
interpolating information from neighboring texels) to cover the
pixel. One pixel maps to a single texel (the ideal case) One pixel maps to several texels. The texel information
must be "minified" to represent the average color of all the covered
texels.
In cases 1 and 3, a MIP map allows a color to be easily approximated
for the screen pixels in near-constant time. This averaging mechanism
minimizes the effects of aliasing as much as possible in a cost-effective
manner. How is a MIP map created? A MIP map is a pyramid of images with the base of the pyramid
equal to the original source texture map. Each successive level
of the texture map is created using a box filter for each texel
in the next level. In other words, four texels in one level of the
map (in a 2×2 square) are averaged to create one texel
in the next, smaller level. Thus, each successive level in a MIP
map is a quarter of area of the previous level. The number of levels
in a MIP map is also referred to as the depth of the map. The highest,
and smallest, level of the MIP map is referred to as the "pinnacle." PEXlib provides the utilities PEXExtCreateFilteredTM
and PEXExtCreateFilteredTMFromWindow
to create a MIP map from a source map provided by the application.
One of the parameters passed to the PEXExtCreateFilteredTM*
utilities is the number of levels to create. If zero is specified,
the optimum number of levels will be created to produce a full MIP
map. How is a MIP map used? The values specified in the Sampling LUT determine how the
MIP map is actually sampled to find the appropriate color for a
single pixel. The minification method specifies how the map should
be sampled when one screen pixel maps to several texels and the
magnification method determines what to do in the case of multiple
pixels mapping to a single texel. The possible methods are detailed
in the table below and vary depending on how many pixels per level
are sampled and how many levels are sampled to determine the final
color. Note that for magnification, only the first two, PEXExtTMTexelSampleSingleBase
and PEXExtTMTexelSampleLinearBase
are recommended. In the table, where "2n"
is specified, n refers to the texture
map dimension: 1, 2, or 3. Table 10-1 MIP
Map Usage Method (minification or magnification) | Texels sampled per level | Level(s) sampled |
|---|
PEXExtTMTexelSampleSingleBase | 1 | Base level only | PEXExtTMTexelSampleLinearBase | 2n | Base level only | PEXExtTMTexelSampleSingleInMipmap | 1 | One level closest to sample depth | PEXExtTMTexelSampleLinearInMipmap | 2n | One level closest to sample depth | PEXExtTMTexelSampleBetweenMipmaps | 1 | Two levels closest to sample depth | PEXExtTMTexelSampleLinearBetweenMipmaps | 2n | Two levels closest to sample depth |
Given these methods, the MIP map is sampled according to the
following formula: The number of texels in the base map
that are covered by a single pixel determines if the minification
method or magnification method should be used. The number of texels
also determines which level should be sampled if the minification
method specifies a level other than the base level. The depth_sampling_bias_hint
value in the Sampling LUT can shift this sampled level up or down
to make the texture appear more detailed or more blurred. The minification method or magnification method,
as appropriate, is used to determine how many pixels should be sampled. If more than one texel is sampled as specified by
the minification method or magnification method, the texels are
averaged to derive the final color value.
User Interface
Considerations for Creating Filtered Texture Maps Some PEXlib implementations require texture map dimensions
to be a power of two and/or require square maps (your implementation's
requirements can be determined by PEXGetImpDepConstants).
The utilities PEXExtCreateFilteredTM
and PEXExtCreateFilteredTMFromWindow
will resize the maps if demanded by the implementation. In some
rare instances, the user may want total control over how the images
are resized to meet the implementation requirements. The user may
want to specify whether the map be cropped to meet the requirements
or shrunk or enlarged using a filter. In these cases, the application
will need to perform the shrinking or enlargement itself, before
calling PEXExtCreateFilteredTM
(or PEXExtCreateFilteredTMFromWindow).
The utilities PEXExtCreateFilteredTM
and PEXExtCreateFilteredTMFromWindow
filter according to these rules: If a power of two is required, each
dimension is resized to the closest power of two. If the closest
power of two is smaller than the original dimension of the map,
the map is down-sampled using a box filter. If the closest power
of two is greater than the original dimension, the map is up-sampled
using linear interpolation. If texture maps are required to be square, the texture
map will be enlarged using linear interpolation to have dimensions
equal to the largest of the two original dimensions. If both power-of-two and square texture maps are
required, the dimensions will first be resized to a power of two
and then the largest dimension will be used as the dimension of
the final map.
Troubleshooting | |
This troubleshooting section is separated into several sections;
choose the desired texture-mapping subject matter. Frequently-Asked Questions - Q.
What areas should be checked if textures fail to
render? - A.
Texture mapping may fail when some of the following
conditions occur: Neither the extended primitives (PEXExtFillAreaSetWithData,
PEXExtSetOfFillAreaSets,
PEXExtTriangleStrip,
PEXExtQuadrilateralMesh)
were called; nor were the OCC alternatives PEXOCCFillArea, PEXOCCFillAreaSet, PEXOCCIndexedFillAreaSets,
PEXOCCTriangleStrip,
and PEXOCCQuadrilateralMesh
called. These are the only primitives for which texture mapping
is supported. The interior style PEXExtInteriorStyleTexture
was not specified for either front and/or back faces. The PEX server does not support the 5.1 texture
mapping extended renderer, determined by inquiring its support with
PEXGetEnumTypeInfo. No texture maps or coordinates were supplied. In
this case, the default map (a black-and-white checkerboard) is used,
but since no texture coordinates existed, the default specifies
a coordinate of (0,0) per vertex, which translates into a single
point of a white square (or black square on some implementations
non-HP implementations). Texture coordinates were out of range when the active
texture coordinate clamp condition was PEXExtTMBoundaryCondBoundary.
Scaling coordinates to be between [0,1) or attempting the clamp
condition PEXExtTMBoundaryCondWrap
(if it is supported) may resolve this. Texture coordinates were out of range when the active
texture coordinate clamp condition was PEXExtTMBoundaryCondClampColor.
If the clamp color specified was black (the default), then the surface
should appear black as well. Try specifying a color of red (1,0,0)
in the Sampling LUT entry to determine if the clamp condition is
being used. The texture map input may be invalid. If so, try
another map type and be certain it abides by any implementation-dependent
constraints such as PEXExtIDPowerOfTwoTMSizesRequired
and PEXExtIDSquareTMRequired.
Note that the utilities PEXExtCreateFilteredTM
and PEXExtCreateFilteredTMFromWindow
can be used to ensure the implementation constants are met. Data from the utilities for texture coordinates
or filtered texture maps was invalid. Be certain to check the return
status of these functions to determine the correctness of their
data.
- Q.
Under what conditions might one see the default
texture (a checkerboard) appear? - A.
The default texture appears when: When the active [backface] texture
list is empty (zero in length). When the active [backface] texture list specifies
a nonexistent Binding LUT entry. When a user Binding LUT table is unspecified. When an invalid TMDescription
or texture map ID is detected.
- Q.
What types of image file formats does PEXlib support? - A.
None. The library supports utilities to perform
xwd-like window
grabs (PEXExtCreateFilteredTMFromWindow)
once an image is displayed, however it cannot read or use a file
format directly. - Q.
Does PEXlib limit the number or size of textures
which can be used? - A.
PEXlib does not limit the number of textures which
can be predefined by PEXExtCreateTM
or PEXExtCreateTMDescription,
however, implementations may limit the actual number of textures
which can be applied to any one extended surface primitive at a
time. This limit is specified by the constant PEXExtIDMaxTextureMaps on
a per-primitive basis. The size of a texture may also be limited
by the value specified in PEXExtIDMaxFastTMSize
if texture mapping acceleration is available to a user. Both values
can be inquired by PEXGetEnumTypeInfo. - Q.
Why do the scaling matrices in the TMCoordSrc LUT behave inversely
to their expected effect on a texture; for example, scaling values
greater than 1 create multiple copies of a map on my surface? - A.
Texture coordinate space is ideally defined between
zero and one [0.0,1.0). When a texture coordinate is scaled by a
value greater than one in the orientation matrix of a TMCoordSrc LUT entry, its
value spans a texture coordinate space distance greater than or
equal to one. In this case, the texture coordinates will, if the
correct wrap or mirror boundary conditions are set, sample more
than one texture "space" in a periodic fashion. For example, the orientation matrix: |3 0 0 0|
|0 2 0 0|
|0 0 1 0|
|0 0 0 1|
transforms the texture coordinates (0,0) and (1,1) to be the
new forms (0,0) and (3,2). As the texture coordinates are interpolated
across texture space, the boundary conditions for zero and one are
examined and the appropriate wrap or clamp actions taken. (To wrap,
compute the value t'0=t0-[t0]
and t'1=t1-[t1].)
In the above example, boundary conditions of PEXExtTMBoundaryCondWrap
for t0 and t1
would result in three copies of the texture in the t0
direction and two copies in the t1 direction.
This stems from the fact that the texture space range of zero to
one [0,1) fits into the above ranges three and two times, respectively. A matrix of the type: |0.25 0 0 0|
| 0 0.25 0 0|
| 0 0 1 0|
| 0 0 0 1|
would map the texture coordinates (0,0) and (1,1) to the new
forms (0,0) and (0.25, 0.25) in texture space, thus sampling only
a fraction of the total map.
- Q.
How does one avoid or fix the unpleasant texture
seams which may occur on an object? - A.
Texture seams can become visible when: The boundary condition PEXExtTMBoundaryCondWrap
is active in a given direction and opposing edges of a texture map
do not match in color. To correct this, ensure the opposing edges
of a map (top=bottom, left=right) share the same color. This can
be performed in many interactive image processing and paint packages.
Alternatively, the boundary condition PEXExtTMBoundaryCondMirror,
if supported, may help in certain situations. Texture coordinates on a primitive do not quite
stretch between [0,1), thus the underlying color of a primitive
(such as white) reveals the edge of the map. In this situation,
explicit texture coordinates can either be scaled and restored in
the vertex to account for this condition (scale operation occurs
once) or the orientation matrix in the TMCoordSrc
LUT can hold a scale matrix (less efficient due to per image cost
to rescale texture coordinates). Two textures in proximity to each other (by nature
of their texture coordinates) do not overlap at their edges cleanly
due to precision problems or because of contrasting colors. Under
these situations, corrections may be possible by either editing
textures to share edge colors explicitly or forcing one texture
to overlap the other via its orientation matrix. The texture coordinate utility functions (PEXExtTMCoord*) are susceptible
to producing parameterizations with obvious "seams." Seams created
here occur when a projection method wraps over a boundary condition
(both cylindrical or spherical projections have seams at their positive
(+X) axes). Using the orientation matrix contained in the param_data argument of these
utility functions can reorient this seam condition to other parts
of an object's geometry (in many desirable cases, the side opposite
the observer).
- Q.
Why does my input texture map appear to be inverted
(upside down) on the textured surface? - A.
The problem may be caused by the ordering of your
texture map scanlines. PEXlib texture maps assume to have their
origins (0,0) at the lower left edge of the quadrant. Many formats
(such as X) assume the upper left. Maps should be inverted in scanline
order if their origins are in conflict with PEXlib. Alternatively,
an appropriate orientation matrix can flip these coordinates as
well, although this method may cause more calculation and precision
costs later in the pipeline. (The utility PEXExtCreateFilteredTMFromWindow
already performs this inversion.)
- Q.
When parameterizing certain objects (such as a sphere
composed of polygons), why do texture maps appear to sometimes "swirl"
around the natural poles of the object? - A.
The poles of a sphere correspond to the "top" and
"bottom" edges of a 2D texture as they collapse down to one point.
Since the texture coordinate utilities PEXExtTMCoord*
compute projections on a per-vertex basis, it is not possible to
represent all points along these edges as one point. Instead, the
projection utility will usually choose one or more points (depending
upon the coordinate source selected), which will then be interpolated
during rendering. When this occurs, a swirling effect appears which
is often emphasized by having large primitives at the top (or bottom)
of the sphere. This visual effect can be minimized by reducing the
sizes of any "polar" primitives of the sphere, but it cannot be
totally eliminated from this parameterization approach. (Ray tracers
escape this problem by evaluating the projection equation at each
pixel, not each vertex, thus minimizing the effect of the polar
singularity.) - Q.
Why do certain primitives (the extended primitives
PEXExtFillAreaSetWithData,
PEXExtSetOfFillAreaSets,
PEXExtTriangleStrip,
and PEXExtQuadrilateralMesh;
or the OCC alternatives PEXOCCFillArea,
PEXOCCFillAreaSet,
PEXOCCIndexedFillAreaSets,
PEXOCCTriangleStrip,
and PEXOCCQuadrilateralMesh)
sometimes appear to have discontinuous textures applied across their
surfaces, even though the texture coordinate utilities worked on
the primitive all at once? - A.
The primitives described above are shared-vertex
types of surface primitives. Although facets of these primitives
may share geometric vertices, they may not actually share the shading
normals associated at these points. Thus, if the texture coordinate
source is PEXExtTMCoordSourceVertexNormal
and a facet normal must be substituted, the coordinates stored in
the vertex will be those of the last shared facet to that vertex
point (which may be incorrect for all other primitives). The only
solutions to this problem are to ensure that facets sharing vertices
also share vertex shading normals when using PEXExtTMCoordSourceVertexNormal
or to use PEXExtTMCoordSourceVertexCoord
when possible. - Q.
What causes textures to become "squeezed" at unnatural
locations on my object geometry? - A.
Spherical projections computed by the texture coordinate
utilities PEXExtTMCoord*
create known artifacts around the poles of their projection object.
These poles lie along the +Y axis in world space (WC). If the natural
axis of symmetry for an object (such as a vase) uses a different
axis (such as the Z axis), then the orientation matrix in the parameterization
data record should incorporate a rotation of 90 degrees to bring
this axis in conjunction with the +Y axis of the projection object.
This will reduce the apparent distortion by keeping the lines of
symmetry aligned.
- Q.
What operations affect the apparent position of
a texture on the surface of a primitive? - A.
The apparent position of a texture on a surface
occurs due to the location of texture coordinates within the 2D
space of a standard texture map. Texture coordinates select regions
of the texture to "pin" to the vertices of a facet. To alter the
selected regions, several measures can be taken: Texture coordinates can be "moved"
during their creation by controlling the method of a surface parameterization.
Both PEXExtTMProjectionSphereWC
and PEXExtTMProjectionCylinderWC
use orientation matrices in their parameterization, effectively
transforming a texture coordinate before it is stored. A TMCoordSrc
LUT entry contains a unique transformation matrix which can scale,
translate, or rotate texture coordinates within 2D texture space.
Use of this feature can aid in exactly positioning texture coordinates
over the interesting areas of a texture map. Editing the contents of the target texture map to
relocate, rescale, or otherwise alter the locations of interest
with respect to the object's texture coordinates.
- Q.
Under certain conditions, why do some of the facets
of a surface fail to texture map when using PEXExtTMParamReflectSphereWC? - A.
The implementation of PEXExtTMParamReflectSphereWC
under HP PEXlib detects a condition where a parameterized primitive
may cross the natural seam of the texture (the seam of the WC sphere
is at the +X axis). To correct this condition, try using the boundary
clamp conditions PEXExtTMBoundaryCondWrap
for both t0 and t1
to ensure correct continuity of the map across a seam. - Q.
What advantage does PEXExtTMParamReflectSphereWC
offer over PEXExtTMParamReflectSphereVRC? - A.
If PEXExtTMParamReflectSphereWC
is supported in your implementation, then reflections on a surface
will change with respect to location of the view in world coordinates
(WC). As an object or viewpoint changes position, so does the apparent
reflection along its surface. This differs from PEXExtTMParamReflectSphereVRC,
which computes reflections in eye coordinates (VRCs), a coordinate
system which always keeps the reflected environment fixed behind
the viewer. From a natural perspective, PEXExtTMParamReflectSphereWC
behaves more like an enclosing environment as the eye position moves
relative to the reflective surface. (HP PEXlib supports the functionality
of PEXExtTMParamReflectSphereWC.) - Q.
How are chrome or metallic effects best simulated? - A.
Chrome and other highly reflective surfaces are
only as interesting as the environments they are found in. In real-world
situations, chrome objects are often photographed in special enclosures
which include both light and dark regions. Texture maps to simulate
this effect should include both light and dark regions of a map
such as those found in sparsely lit rooms. Photographs or other
image data can be converted specifically for this purpose. The quality
of rendered reflection is also dependent upon the clarity of detail
in a texture map. The sharper the detail contained in a map, the
greater the impression of fine chrome there will be on a rendered
model.
- Q.
Under what conditions will my texture mapping performance
be the best? - A.
Several steps can be taken to improve texture mapping
throughput in the system: Attempt to render all extended surface
primitives using the same set of textures at the same time. Activating
different textures on a per-primitive basis can be costly. Minimize the mixture of surface primitives going
to PEX. When the supported extended primitives for texture mapping
(PEXExtFillAreaSetWithData,
PEXExtSetOfFillAreaSets,
PEXExtTriangleStrip,
PEXExtQuadrilateralMesh),
or the OCC versions of these primitives (PEXOCCFillArea, PEXOCCFillAreaSet, PEXOCCIndexedFillAreaSets,
PEXOCCTriangleStrip,
and PEXOCCQuadrilateralMesh)
are mixed with other types of primitives, the overall cost to switch
modes — texturing versus non-texturing — will
increase. A single active texture map will generally perform
better than multiple active texture maps. The performance cost,
however, varies with general size of the map, type of map, number
of maps, and the complexity of the texture operations associated
with each map. Texture maps which use the "Int8" (byte) level organization
save considerable memory resource on a given system. Larger maps,
because of their dimension or the fact that they are comprised of
"Float" data,
will impact the operating system by increasing the application's
use of swap space. Pre-transforming texture coordinate data stored
at a vertex and using identity orientation matrices for all TMCoordSrc LUT entries will
reduce the amount of computation spent during texture mapping. Although
these matrices help position a texture on the surface of an object,
it is often possible to fix these texture coordinate values by pre-transforming
them one time since they are not likely to change after their location
can be tied down.
- Q.
My texture map has many fine details in it, some
only one texel wide (such as lines). When I rotate the surface it
is applied to, these lines tend to break up or flash. What can I
do to correct this? - A.
Texture mapping is all about sampling theory. Essentially,
the more discrete samples taken, the better the reconstruction of
the original, continuous signal. The artifact you are experiencing
is sampling "aliasing," the loss of reconstruction information.
To reduce or eliminate visual artifacts, try the following: Increase the size of your texture
map. Since aliasing is based on a factor of screen pixel size to
its coverage on the texture map, a greater size map with more information
can often improve color sampling. Use a MIP map with more than one level. MIP maps
are precomputed sampling filters which try to account for aliasing
during sampling. Use of a MIP map can often improve image quality
during animation due to problems with sampling. Use better sampling methods for your MIP map. PEXExtTMTexelSampleSingleBase
uses only one sample and is prone to aliasing. PEXExtTMTexelSampleLinearBetweenMipmaps
uses many samples and an neighbor interpolation filter to approximate
colors, thus reducing the inaccuracy of single point sampling. (The
only drawback to using more interpolation is a softening of sharp
detail due to the blending used in the computation.)
- Q.
Portions of my textured surface appear to be blotched
with regions of blurred detail. What are they and how can I correct
this? - A.
The artifact witnessed here occurs due to the design
of the Mip Map anti-aliasing filter. As a renderer attempts to approximate
an area of texels to sample in texture space, it selects different
map levels to sample. The regions of blurred detail are visible
transitions which are occurring due to the current sampling method.
To reduce the visual impacts of these artifacts, several steps should
be attempted: Create your own MIP maps using better
filters than those offered by PEXExtCreateFilteredTM
and PEXExtCreateFilteredTMFromWindow.
These utilities employ box- and linear filters to minify and scale
incoming image data. Other image processing filters such as Gaussian
can lead to improved images. Increase the level of filtering. PEXExtTMTexelSampleLinearBetweenMipmaps
provides the highest quality of visual control, however, it also
requires the greatest amount of computation; this can reduce pipeline
throughput. Increase the size of the depth_sampling_bias_hint
in the TMSampling
LUT to greater than one to force sampling to occur in deeper maps
away from the base map. The overall blurring of detail will increase,
however it will appear more uniform. Decrease the size to improve
sharpness. Decrease the t*_frequency_hint in any of
the t0 or t1
directions. Depending upon the level of detail in a map and its
frequency, this bias will reduce the contribution of t0
or t1 gradient calculations such that
the levels of Mip Map will change more rapidly based upon the contributing
factor of the other's (t1 or t0)
gradient. This hint, however, may not be implemented under all implementations
or devices. Reduce the level of high frequency detail in the
original texture map. Areas of high change (gradient) reveal greater
levels of visual artifacts when viewed on a textured surface. Decrease the size of the primitives to which the
texture will be applied. Sometimes the interaction between light
shade interpolation and texture surface causes mach-bands or other
artifacts to become more pronounced. Increasing the level of detail
can sometimes improve the final visual outcome.
|
Printable version
