mitsuba/src/sensors/perspective.cpp

477 lines
16 KiB
C++

/*
This file is part of Mitsuba, a physically based rendering system.
Copyright (c) 2007-2012 by Wenzel Jakob and others.
Mitsuba is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License Version 3
as published by the Free Software Foundation.
Mitsuba is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include <mitsuba/render/sensor.h>
#include <mitsuba/render/medium.h>
#include <mitsuba/core/track.h>
#include <mitsuba/core/frame.h>
MTS_NAMESPACE_BEGIN
/*!\plugin{perspective}{Perspective pinhole camera}
* \order{1}
* \parameters{
* \parameter{toWorld}{\Transform\Or\ATransform}{
* Specifies an optional camera-to-world transformation.
* \default{none (i.e. camera space $=$ world space)}
* }
* \parameter{focalLength}{\String}{
* Denotes the camera's focal length specified using
* \code{35mm} film equivalent units. See the main
* description for further details.
* \default{\code{50mm}}
* }
* \parameter{fov}{\Float}{
* An alternative to \code{focalLength}:
* denotes the camera's field of view in degrees---must be
* between 0 and 180, excluding the extremes.
* }
* \parameter{fovAxis}{\String}{
* When the parameter \code{fov} is given (and only then),
* this parameter further specifies the image axis, to
* which it applies.
* \begin{enumerate}[(i)]
* \item \code{\textbf{x}}: \code{fov} maps to the
* \code{x}-axis in screen space.
* \item \code{\textbf{y}}: \code{fov} maps to the
* \code{y}-axis in screen space.
* \item \code{\textbf{diagonal}}: \code{fov}
* maps to the screen diagonal.
* \item \code{\textbf{smaller}}: \code{fov}
* maps to the smaller dimension
* (e.g. \code{x} when \code{width}<\code{height})
* \item \code{\textbf{larger}}: \code{fov}
* maps to the larger dimension
* (e.g. \code{y} when \code{width}<\code{height})
* \end{enumerate}
* The default is \code{\textbf{x}}.
* }
* \parameter{shutterOpen, shutterClose}{\Float}{
* Specifies the time interval of the measurement---this
* is only relevant when the scene is in motion.
* \default{0}
* }
* \parameter{nearClip, farClip}{\Float}{
* Distance to the near/far clip
* planes.\default{\code{near\code}-\code{Clip=1e-2} (i.e.
* \code{0.01}) and {\code{farClip=1e4} (i.e. \code{10000})}}
* }
* }
* \renderings{
* \rendering{The material test ball viewed through a perspective pinhole
* camera. Everything is in sharp focus.}{sensor_perspective}
* \medrendering{A rendering of the Cornell box}{sensor_perspective_2}
* }
*
* This plugin implements a simple idealizied perspective camera model, which
* has an infinitely small aperture. This creates an infinite depth of field,
* i.e. no optical blurring occurs. The camera is can be specified to move during
* an exposure, hence temporal blur is still possible.
*
* By default, the camera's field of view is specified using a 35mm film
* equivalent focal length, which is first converted into a diagonal field
* of view and subsequently applied to the camera. This assumes that
* the film's aspect ratio matches that of 35mm film (1.5:1), though the
* parameter still behaves intuitively when this is not the case.
* Alternatively, it is also possible to specify a field of view in degrees
* along a given axis (see the \code{fov} and \code{fovAxis} parameters).
*
* The exact camera position and orientation is most easily expressed using the
* \code{lookat} tag, i.e.:
* \begin{xml}
* <sensor type="perspective">
* <transform name="toWorld">
* <!-- Move and rotate the camera so that looks from (1, 1, 1) to (1, 2, 1)
* and the direction (0, 0, 1) points "up" in the output image -->
* <lookat origin="1, 1, 1" target="1, 2, 1" up="0, 0, 1"/>
* </transform>
* </sensor>
* \end{xml}
*/
class PerspectiveCameraImpl : public PerspectiveCamera {
public:
PerspectiveCameraImpl(const Properties &props)
: PerspectiveCamera(props) {
/* This sensor is the result of a limiting process where the aperture
radius tends to zero. However, it still has all the cosine
foreshortening terms caused by the aperture, hence the flag "EOnSurface" */
m_type |= EDeltaPosition | EPerspectiveCamera | EOnSurface | EDirectionSampleMapsToPixels;
if (props.getAnimatedTransform("toWorld", Transform())->eval(0).hasScale())
Log(EError, "Scale factors in the camera-to-world "
"transformation are not allowed!");
}
PerspectiveCameraImpl(Stream *stream, InstanceManager *manager)
: PerspectiveCamera(stream, manager) {
configure();
}
void configure() {
PerspectiveCamera::configure();
const Vector2i &filmSize = m_film->getSize();
const Vector2i &cropSize = m_film->getCropSize();
const Point2i &cropOffset = m_film->getCropOffset();
Vector2 relSize((Float) cropSize.x / (Float) filmSize.x,
(Float) cropSize.y / (Float) filmSize.y);
Point2 relOffset((Float) cropOffset.x / (Float) filmSize.x,
(Float) cropOffset.y / (Float) filmSize.y);
/**
* These do the following (in reverse order):
*
* 1. Create transform from camera space to [-1,1]x[-1,1]x[0,1] clip
* coordinates (not taking account of the aspect ratio yet)
*
* 2+3. Translate and scale to shift the clip coordinates into the
* range from zero to one, and take the aspect ratio into account.
*
* 4+5. Translate and scale the coordinates once more to account
* for a cropping window (if there is any)
*/
m_cameraToSample =
Transform::scale(Vector(1.0f / relSize.x, 1.0f / relSize.y, 1.0f))
* Transform::translate(Vector(-relOffset.x, -relOffset.y, 0.0f))
* Transform::scale(Vector(-0.5f, -0.5f*m_aspect, 1.0f))
* Transform::translate(Vector(-1.0f, -1.0f/m_aspect, 0.0f))
* Transform::perspective(m_xfov, m_nearClip, m_farClip);
m_sampleToCamera = m_cameraToSample.inverse();
/* Position differentials on the near plane */
m_dx = m_sampleToCamera(Point(m_invResolution.x, 0.0f, 0.0f))
- m_sampleToCamera(Point(0.0f));
m_dy = m_sampleToCamera(Point(0.0f, m_invResolution.y, 0.0f))
- m_sampleToCamera(Point(0.0f));
/* Precompute some data for importance(). Please
look at that function for further details */
Point min(m_sampleToCamera(Point(0, 0, 0))),
max(m_sampleToCamera(Point(1, 1, 0)));
m_imageRect.reset();
m_imageRect.expandBy(Point2(min.x, min.y) / min.z);
m_imageRect.expandBy(Point2(max.x, max.y) / max.z);
m_normalization = 1.0f / m_imageRect.getVolume();
/* Clip-space transformation for OpenGL */
m_clipTransform = Transform::translate(
Vector((1-2*relOffset.x)/relSize.x - 1,
-(1-2*relOffset.y)/relSize.y + 1, 0.0f)) *
Transform::scale(Vector(1.0f / relSize.x, 1.0f / relSize.y, 1.0f));
}
/**
* \brief Compute the directional sensor response function
* of the camera multiplied with the cosine foreshortening
* factor associated with the image plane
*
* \param d
* A normalized direction vector from the aperture position to the
* reference point in question (all in local camera space)
*/
inline Float importance(const Vector &d) const {
/* How is this derived? Imagine a hypothetical image plane at a
distance of d=1 away from the pinhole in camera space.
Then the visible rectangular portion of the plane has the area
A = (2 * tan(0.5 * xfov in radians))^2 / aspect
Since we allow crop regions, the actual visible area is
potentially reduced:
A' = A * (cropX / filmX) * (cropY / filmY)
Perspective transformations of such aligned rectangles produce
an equivalent scaled (but otherwise undistorted) rectangle
in screen space. This means that a strategy, which uniformly
generates samples in screen space has an associated area
density of 1/A' on this rectangle.
To compute the solid angle density of a sampled point P on
the rectangle, we can apply the usual measure conversion term:
d_omega = 1/A' * distance(P, origin)^2 / cos(theta)
where theta is the angle that the unit direction vector from
the origin to P makes with the rectangle. Since
distance(P, origin)^2 = Px^2 + Py^2 + 1
and
cos(theta) = 1/sqrt(Px^2 + Py^2 + 1),
we have
d_omega = 1 / (A' * cos^3(theta))
*/
Float cosTheta = Frame::cosTheta(d);
/* Check if the direction points behind the camera */
if (cosTheta <= 0)
return 0.0f;
/* Compute the position on the plane at distance 1 */
Float invCosTheta = 1.0f / cosTheta;
Point2 p(d.x * invCosTheta, d.y * invCosTheta);
/* Check if the point lies inside the chosen crop rectangle */
if (!m_imageRect.contains(p))
return 0.0f;
return m_normalization * invCosTheta
* invCosTheta * invCosTheta;
}
Spectrum sampleRay(Ray &ray, const Point2 &pixelSample,
const Point2 &otherSample, Float timeSample) const {
ray.time = sampleTime(timeSample);
/* Compute the corresponding position on the
near plane (in local camera space) */
Point nearP = m_sampleToCamera(Point(
pixelSample.x * m_invResolution.x,
pixelSample.y * m_invResolution.y, 0.0f));
/* Turn that into a normalized ray direction, and
adjust the ray interval accordingly */
Vector d = normalize(Vector(nearP));
Float invZ = 1.0f / d.z;
ray.mint = m_nearClip * invZ;
ray.maxt = m_farClip * invZ;
const Transform &trafo = m_worldTransform->eval(ray.time);
ray.setOrigin(trafo.transformAffine(Point(0.0f)));
ray.setDirection(trafo(d));
return Spectrum(1.0f);
}
Spectrum sampleRayDifferential(RayDifferential &ray, const Point2 &pixelSample,
const Point2 &otherSample, Float timeSample) const {
ray.time = sampleTime(timeSample);
/* Compute the corresponding position on the
near plane (in local camera space) */
Point nearP = m_sampleToCamera(Point(
pixelSample.x * m_invResolution.x,
pixelSample.y * m_invResolution.y, 0.0f));
/* Turn that into a normalized ray direction, and
adjust the ray interval accordingly */
Vector d = normalize(Vector(nearP));
Float invZ = 1.0f / d.z;
ray.mint = m_nearClip * invZ;
ray.maxt = m_farClip * invZ;
const Transform &trafo = m_worldTransform->eval(ray.time);
ray.setOrigin(trafo.transformAffine(Point(0.0f)));
ray.setDirection(trafo(d));
ray.rxOrigin = ray.ryOrigin = ray.o;
ray.rxDirection = trafo(normalize(Vector(nearP) + m_dx));
ray.ryDirection = trafo(normalize(Vector(nearP) + m_dy));
ray.hasDifferentials = true;
return Spectrum(1.0f);
}
Spectrum samplePosition(PositionSamplingRecord &pRec,
const Point2 &sample, const Point2 *extra) const {
const Transform &trafo = m_worldTransform->eval(pRec.time);
pRec.p = trafo(Point(0.0f));
pRec.n = trafo(Vector(0.0f, 0.0f, 1.0f));
pRec.pdf = 1.0f;
pRec.measure = EDiscrete;
return Spectrum(1.0f);
}
Spectrum evalPosition(const PositionSamplingRecord &pRec) const {
return Spectrum((pRec.measure == EDiscrete) ? 1.0f : 0.0f);
}
Float pdfPosition(const PositionSamplingRecord &pRec) const {
return (pRec.measure == EDiscrete) ? 1.0f : 0.0f;
}
Spectrum sampleDirection(DirectionSamplingRecord &dRec,
PositionSamplingRecord &pRec,
const Point2 &sample, const Point2 *extra) const {
const Transform &trafo = m_worldTransform->eval(pRec.time);
Point samplePos(sample.x, sample.y, 0.0f);
if (extra) {
/* The caller wants to condition on a specific pixel position */
samplePos.x = (extra->x + sample.x) * m_invResolution.x;
samplePos.y = (extra->y + sample.y) * m_invResolution.y;
}
pRec.uv = Point2(samplePos.x * m_resolution.x,
samplePos.y * m_resolution.y);
/* Compute the corresponding position on the
near plane (in local camera space) */
Point nearP = m_sampleToCamera(samplePos);
/* Turn that into a normalized ray direction */
Vector d = normalize(Vector(nearP));
dRec.d = trafo(d);
dRec.measure = ESolidAngle;
dRec.pdf = m_normalization / (d.z * d.z * d.z);
return Spectrum(1.0f);
}
Float pdfDirection(const DirectionSamplingRecord &dRec,
const PositionSamplingRecord &pRec) const {
if (dRec.measure != ESolidAngle)
return 0.0f;
const Transform &trafo = m_worldTransform->eval(pRec.time);
return importance(trafo.inverse()(dRec.d));
}
Spectrum evalDirection(const DirectionSamplingRecord &dRec,
const PositionSamplingRecord &pRec) const {
if (dRec.measure != ESolidAngle)
return Spectrum(0.0f);
const Transform &trafo = m_worldTransform->eval(pRec.time);
return Spectrum(importance(trafo.inverse()(dRec.d)));
}
bool getSamplePosition(const PositionSamplingRecord &pRec,
const DirectionSamplingRecord &dRec, Point2 &samplePosition) const {
Transform invTrafo = m_worldTransform->eval(pRec.time).inverse();
Point local(Point(invTrafo(dRec.d)));
if (local.z <= 0)
return false;
Point screenSample = m_cameraToSample(local);
if (screenSample.x < 0 || screenSample.x > 1 ||
screenSample.y < 0 || screenSample.y > 1)
return false;
samplePosition = Point2(
screenSample.x * m_resolution.x,
screenSample.y * m_resolution.y);
return true;
}
Spectrum sampleDirect(DirectSamplingRecord &dRec, const Point2 &sample) const {
const Transform &trafo = m_worldTransform->eval(dRec.time);
/* Transform the reference point into the local coordinate system */
Point refP = trafo.inverse().transformAffine(dRec.ref);
/* Check if it is outside of the clip range */
if (refP.z < m_nearClip || refP.z > m_farClip) {
dRec.pdf = 0.0f;
return Spectrum(0.0f);
}
Point screenSample = m_cameraToSample(refP);
dRec.uv = Point2(screenSample.x, screenSample.y);
if (dRec.uv.x < 0 || dRec.uv.x > 1 ||
dRec.uv.y < 0 || dRec.uv.y > 1) {
dRec.pdf = 0.0f;
return Spectrum(0.0f);
}
dRec.uv.x *= m_resolution.x;
dRec.uv.y *= m_resolution.y;
Vector localD(refP);
Float dist = localD.length(),
invDist = 1.0f / dist;
localD *= invDist;
dRec.p = trafo.transformAffine(Point(0.0f));
dRec.d = (dRec.p - dRec.ref) * invDist;
dRec.dist = dist;
dRec.n = trafo(Vector(0.0f, 0.0f, 1.0f));
dRec.pdf = 1;
dRec.measure = EDiscrete;
return Spectrum(
importance(localD) * invDist * invDist);
}
Float pdfDirect(const DirectSamplingRecord &dRec) const {
return (dRec.measure == EDiscrete) ? 1.0f : 0.0f;
}
Transform getProjectionTransform(const Point2 &apertureSample,
const Point2 &aaSample) const {
Float right = std::tan(m_xfov * M_PI/360) * m_nearClip, left = -right;
Float top = right / m_aspect, bottom = -top;
Vector2 offset(
(right-left)/m_film->getSize().x * (aaSample.x-0.5f),
(top-bottom)/m_film->getSize().y * (aaSample.y-0.5f));
return m_clipTransform *
Transform::glFrustum(left+offset.x, right+offset.x,
bottom+offset.y, top+offset.y, m_nearClip, m_farClip);
}
AABB getAABB() const {
return m_worldTransform->getTranslationBounds();
}
std::string toString() const {
std::ostringstream oss;
oss << "PerspectiveCamera[" << endl
<< " xfov = " << m_xfov << "," << endl
<< " nearClip = " << m_nearClip << "," << endl
<< " farClip = " << m_farClip << "," << endl
<< " worldTransform = " << indent(m_worldTransform.toString()) << "," << endl
<< " sampler = " << indent(m_sampler->toString()) << "," << endl
<< " film = " << indent(m_film->toString()) << "," << endl
<< " medium = " << indent(m_medium.toString()) << "," << endl
<< " shutterOpen = " << m_shutterOpen << "," << endl
<< " shutterOpenTime = " << m_shutterOpenTime << endl
<< "]";
return oss.str();
}
MTS_DECLARE_CLASS()
private:
Transform m_cameraToSample;
Transform m_sampleToCamera;
Transform m_clipTransform;
AABB2 m_imageRect;
Float m_normalization;
Vector m_dx, m_dy;
};
MTS_IMPLEMENT_CLASS_S(PerspectiveCameraImpl, false, PerspectiveCamera)
MTS_EXPORT_PLUGIN(PerspectiveCameraImpl, "Perspective camera");
MTS_NAMESPACE_END