void ReadFile(CImage& img, const char* filename)
{
// Determine the file extension
const char *dot = strrchr(filename, '.');
if (strcasecmp(dot, ".tga") == 0 || strcasecmp(dot, ".tga") == 0) {
if ((&img.PixType()) == 0)
img.ReAllocate(CShape(), typeid(uchar), sizeof(uchar), true);
if (img.PixType() == typeid(uchar))
ReadFileTGA(*(CByteImage *) &img, filename);
else {
CByteImage imgAux;
ReadFileTGA(imgAux, filename);
if(img.PixType() == typeid(float)) TypeConvert(imgAux, *(CFloatImage*) &img);
else
throw CError("Cannot load image into img with this type of buffer");
}
} else if (strcasecmp(dot, ".jpg") == 0 || strcasecmp(dot, ".jpeg") == 0) {
if ((&img.PixType()) == 0)
img.ReAllocate(CShape(), typeid(uchar), sizeof(uchar), true);
if (img.PixType() == typeid(uchar))
ReadFileJPEG(*(CByteImage *) &img, filename);
else {
CByteImage imgAux;
ReadFileJPEG(imgAux, filename);
if(img.PixType() == typeid(float)) TypeConvert(imgAux, *(CFloatImage*) &img);
else
throw CError("Cannot load image into img with this type of buffer");
}
} else
throw CError("ReadFile(%s): file type not supported", filename);
}
CByteImage
HOGFeatureExtractor::render(const Feature& f) const
{
CShape cellShape = _oriMarkers[0].Shape();
CFloatImage hogImgF(CShape(cellShape.width * f.Shape().width, cellShape.height * f.Shape().height, 1));
hogImgF.ClearPixels();
float minBinValue, maxBinValue;
f.getRangeOfValues(minBinValue, maxBinValue);
// For every cell in the HOG
for(int hi = 0; hi < f.Shape().height; hi++) {
for(int hj = 0; hj < f.Shape().width; hj++) {
// Now _oriMarkers, multiplying contribution by bin level
for(int hc = 0; hc < _nAngularBins; hc++) {
float v = f.Pixel(hj, hi, hc) / maxBinValue;
for(int ci = 0; ci < cellShape.height; ci++) {
float* cellIt = (float*) _oriMarkers[hc].PixelAddress(0, ci, 0);
float* hogIt = (float*) hogImgF.PixelAddress(hj * cellShape.height, hi * cellShape.height + ci, 0);
for(int cj = 0; cj < cellShape.width; cj++, hogIt++, cellIt++) {
(*hogIt) += v * (*cellIt);
}
}
}
}
}
CByteImage hogImg;
TypeConvert(hogImgF, hogImg);
return hogImg;
}
KernelInit::KernelInit()
{
static float k_11[2] = {0.5f, 0.5f};
static float k_121[3] = {0.25f, 0.5f, 0.25f};
static float k_14641[5] = {0.0625f, 0.25f, 0.375f, 0.25f, 0.0625f};
static float k_8ptFP[8] = {-0.044734f, -0.059009f, 0.156544f, 0.449199f,
0.449199f, 0.156544f, -0.059009f, -0.044734f};
// The following are derived as fix-point /256 fractions of the above:
// -12, -15, 40, 115
static float k_8ptI [8] = {-0.04687500f, -0.05859375f, 0.15625000f, 0.44921875f,
0.44921875f, 0.15625000f, -0.05859375f, -0.04687500f};
ConvolveKernel_121.ReAllocate(CShape(3, 1, 1), k_121, false, 3);
ConvolveKernel_121.origin[0] = -1;
ConvolveKernel_14641.ReAllocate(CShape(5, 1, 1), k_14641, false, 5);
ConvolveKernel_14641.origin[0] = -2;
ConvolveKernel_8tapLowPass.ReAllocate(CShape(8, 1, 1), k_8ptI, false, 8);
ConvolveKernel_8tapLowPass.origin[0] = -4;
}
cUFO::cUFO(double x, double y, TGame* game)
: CMoveAbleObject(CShape(
new CPrimaryShape(SHAPE_TYPE_OVAL, x + 44, y, 70, 60),
new CPrimaryShape(SHAPE_TYPE_OVAL, x+0, y+50, 150, 50)), UFO, 250.0, 20.0, 0.0)
{
beamZone = new CPrimaryShape(SHAPE_TYPE_RECT, x, y + 100, 150, 50);
this->game = game;
this->setVelX(UFO_VELX);
animation.setLastAnimationTime(SDL_GetTicks() - UFO_BEAM_TIME_DIFFERENCE);
}
void Convolve(CImageOf<T> src, CImageOf<T>& dst,
CFloatImage kernel,
float scale, float offset)
{
// Determine the shape of the kernel and row buffer
CShape kShape = kernel.Shape();
CShape sShape = src.Shape();
CShape bShape(sShape.width + kShape.width, kShape.height, sShape.nBands);
int bWidth = bShape.width * bShape.nBands;
// Allocate the result, if necessary, and the row buffer
dst.ReAllocate(sShape, false);
CFloatImage buffer(bShape);
if (sShape.width * sShape.height * sShape.nBands == 0)
return;
CFloatImage output(CShape(sShape.width, 1, sShape.nBands));
// Fill up the row buffer initially
for (int k = 0; k < kShape.height; k++)
FillRowBuffer(&buffer.Pixel(0, k, 0), src, kernel, k, bWidth);
// Determine if clipping is required
// (we assume up-conversion to float never requires clipping, i.e.,
// floats have the highest dynamic range)
T minVal = dst.MinVal();
T maxVal = dst.MaxVal();
if (minVal <= buffer.MinVal() && maxVal >= buffer.MaxVal())
minVal = maxVal = 0;
// Process each row
for (int y = 0; y < sShape.height; y++)
{
// Do the convolution
ConvolveRow2D(buffer, kernel, &output.Pixel(0, 0, 0),
sShape.width);
// Scale, offset, and type convert
ScaleAndOffsetLine(&output.Pixel(0, 0, 0), &dst.Pixel(0, y, 0),
sShape.width * sShape.nBands,
scale, offset, minVal, maxVal);
// Shift up the row buffer and fill the last line
if (y < sShape.height-1)
{
int k;
for (k = 0; k < kShape.height-1; k++)
memcpy(&buffer.Pixel(0, k, 0), &buffer.Pixel(0, k+1, 0),
bWidth * sizeof(float));
FillRowBuffer(&buffer.Pixel(0, k, 0), src, kernel, y+k+1, bWidth);
}
}
}
void MotionToColor(CFloatImage motim, CByteImage &colim, float maxmotion)
{
CShape sh = motim.Shape();
int width = sh.width, height = sh.height;
colim.ReAllocate(CShape(width, height, 3));
int x, y;
// determine motion range:
float maxx = -999, maxy = -999;
float minx = 999, miny = 999;
float maxrad = -1;
for (y = 0; y < height; y++) {
for (x = 0; x < width; x++) {
float fx = motim.Pixel(x, y, 0);
float fy = motim.Pixel(x, y, 1);
if (unknown_flow(fx, fy))
continue;
maxx = __max(maxx, fx);
maxy = __max(maxy, fy);
minx = __min(minx, fx);
miny = __min(miny, fy);
float rad = sqrt(fx * fx + fy * fy);
maxrad = __max(maxrad, rad);
}
}
printf("max motion: %.4f motion range: u = %.3f .. %.3f; v = %.3f .. %.3f\n",
maxrad, minx, maxx, miny, maxy);
if (maxmotion > 0) // i.e., specified on commandline
maxrad = maxmotion;
if (maxrad == 0) // if flow == 0 everywhere
maxrad = 1;
if (verbose)
fprintf(stderr, "normalizing by %g\n", maxrad);
for (y = 0; y < height; y++) {
for (x = 0; x < width; x++) {
float fx = motim.Pixel(x, y, 0);
float fy = motim.Pixel(x, y, 1);
uchar *pix = &colim.Pixel(x, y, 0);
if (unknown_flow(fx, fy)) {
pix[0] = pix[1] = pix[2] = 0;
} else {
computeColor(fx/maxrad, fy/maxrad, pix);
}
}
}
}
void ReadFile (CImage& img, const char* filename)
{
// Determine the file extension
char *dot = strrchr((char *) filename, '.');
if (strcmp(dot, ".tga") == 0 || strcmp(dot, ".tga") == 0)
{
if ((&img.PixType()) == 0)
img.ReAllocate(CShape(), typeid(uchar), sizeof(uchar), true);
if (img.PixType() == typeid(uchar))
ReadFileTGA(*(CByteImage *) &img, filename);
else
throw CError("ReadFile(%s): haven't implemented conversions yet", filename);
}
else
throw CError("ReadFile(%s): file type not supported", filename);
}
KernelInit::KernelInit()
{
static float k_11[2] = {0.5f, 0.5f};
static float k_121[3] = {0.25f, 0.5f, 0.25f};
static float k_14641[5] = {0.0625f, 0.25f, 0.375f, 0.25f, 0.0625f};
static float k_8ptFP[8] = {-0.044734f, -0.059009f, 0.156544f, 0.449199f,
0.449199f, 0.156544f, -0.059009f, -0.044734f};
// The following are derived as fix-point /256 fractions of the above:
// -12, -15, 40, 115
static float k_8ptI [8] = {-0.04687500f, -0.05859375f, 0.15625000f, 0.44921875f,
0.44921875f, 0.15625000f, -0.05859375f, -0.04687500f};
static float k_7x7[49] = { 1.0, 4.0, 7.0, 10.0, 7.0, 4.0, 1.0,
4.0, 12.0, 26.0, 33.0, 26.0, 12.0, 4.0,
7.0, 26.0, 55.0, 71.0, 55.0, 26.0, 7.0,
10.0, 33.0, 71.0, 91.0, 71.0, 33.0, 10.0,
7.0, 26.0, 55.0, 71.0, 55.0, 26.0, 7.0,
4.0, 12.0, 26.0, 33.0, 26.0, 12.0, 4.0,
1.0, 4.0, 7.0, 10.0, 7.0, 4.0, 1.0 };
for (int i = 0; i < 49; i++) {
k_7x7[i] /= 1115.0;
}
ConvolveKernel_121.ReAllocate(CShape(3, 1, 1), k_121, false, 3);
ConvolveKernel_121.origin[0] = 1;
ConvolveKernel_14641.ReAllocate(CShape(5, 1, 1), k_14641, false, 5);
ConvolveKernel_14641.origin[0] = 2;
ConvolveKernel_8tapLowPass.ReAllocate(CShape(8, 1, 1), k_8ptI, false, 8);
ConvolveKernel_8tapLowPass.origin[0] = 4;
ConvolveKernel_7x7.ReAllocate(CShape(7, 7, 1), k_7x7, false, 7);
/* Sobel filters */
static float k_SobelX[9] = { -1, 0, 1,
-2, 0, 2,
-1, 0, 1 };
static float k_SobelY[9] = { -1, -2, -1,
0, 0, 0,
1, 2, 1 };
ConvolveKernel_SobelX.ReAllocate(CShape(3, 3, 1), k_SobelX, false, 3);
ConvolveKernel_SobelX.origin[0] = 1;
ConvolveKernel_SobelX.origin[1] = 1;
ConvolveKernel_SobelY.ReAllocate(CShape(3, 3, 1), k_SobelY, false, 3);
ConvolveKernel_SobelY.origin[0] = 1;
ConvolveKernel_SobelY.origin[1] = 1;
}
void
SupportVectorMachine::predictSlidingWindow(const Feature &feat, CFloatImage &response) const
{
response.ReAllocate(CShape(feat.Shape().width, feat.Shape().height, 1));
response.ClearPixels();
/******** BEGIN TODO ********/
// Sliding window prediction.
//
// In this project we are using a linear SVM. This means that
// it's classification function is very simple, consisting of a
// dot product of the feature vector with a set of weights learned
// during training, followed by a subtraction of a bias term
//
// pred <- dot(feat, weights) - bias term
//
// Now this is very simple to compute when we are dealing with
// cropped images, our computed features have the same dimensions
// as the SVM weights. Things get a little more tricky when you
// want to evaluate this function over all possible subwindows of
// a larger feature, one that we would get by running our feature
// extraction on an entire image.
//
// Here you will evaluate the above expression by breaking
// the dot product into a series of convolutions (remember that
// a convolution can be though of as a point wise dot product with
// the convolution kernel), each one with a different band.
//
// Convolve each band of the SVM weights with the corresponding
// band in feat, and add the resulting score image. The final
// step is to subtract the SVM bias term given by this->getBiasTerm().
//
// Hint: you might need to set the origin for the convolution kernel
// in order to get the result from convoltion to be correctly centered
//
// Useful functions:
// Convolve, BandSelect, this->getWeights(), this->getBiasTerm()
Feature weights = this->getWeights();
int nWtBands = weights.Shape().nBands;
// Set the center of the window as the origin for the conv. kernel
for (int band = 0; band < nWtBands; band++)
{
// Select a band
CFloatImage featBand;
CFloatImage weightBand;
BandSelect(feat, featBand, band, 0);
BandSelect(weights, weightBand, band, 0);
// Set the origin of the kernel
weightBand.origin[0] = weights.Shape().width / 2;
weightBand.origin[1] = weights.Shape().height / 2;
// Compute the dot product
CFloatImage dotproduct;
dotproduct.ClearPixels();
Convolve(featBand, dotproduct, weightBand);
// Add the resulting score image
for (int y = 0; y < feat.Shape().height; y++)
{
for (int x = 0; x < feat.Shape().width; x++)
{
response.Pixel(x, y, 0) += dotproduct.Pixel(x, y, 0);
}
// End of x loop
}
// End of y loop
}
// End of band loop
// Substract the SVM bias term
for (int y = 0; y < feat.Shape().height; y++)
{
for (int x = 0; x < feat.Shape().width; x++)
{
response.Pixel(x, y, 0) -= this->getBiasTerm();
}
// End of x loop
}
// End of y loop
/******** END TODO ********/
}
void CImage::DeAllocate()
{
// Release the memory & set to default values
ReAllocate(CShape(), *(const type_info *) 0, 0, 0, false, 0);
SetDefaults();
}
CFloatImage
SupportVectorMachine::predictSlidingWindow(const Feature& feat) const
{
CFloatImage score(CShape(feat.Shape().width,feat.Shape().height,1));
score.ClearPixels();
/******** BEGIN TODO ********/
// Sliding window prediction.
//
// In this project we are using a linear SVM. This means that
// it's classification function is very simple, consisting of a
// dot product of the feature vector with a set of weights learned
// during training, followed by a subtraction of a bias term
//
// pred <- dot(feat, weights) - bias term
//
// Now this is very simple to compute when we are dealing with
// cropped images, our computed features have the same dimensions
// as the SVM weights. Things get a little more tricky when you
// want to evaluate this function over all possible subwindows of
// a larger feature, one that we would get by running our feature
// extraction on an entire image.
//
// Here you will evaluate the above expression by breaking
// the dot product into a series of convolutions (remember that
// a convolution can be though of as a point wise dot product with
// the convolution kernel), each one with a different band.
//
// Convolve each band of the SVM weights with the corresponding
// band in feat, and add the resulting score image. The final
// step is to subtract the SVM bias term given by this->getBiasTerm().
//
// Hint: you might need to set the origin for the convolution kernel
// in order to get the result from convoltion to be correctly centered
//
// Useful functions:
// Convolve, BandSelect, this->getWeights(), this->getBiasTerm()
//printf("TODO: SupportVectorMachine.cpp:273\n"); exit(EXIT_FAILURE);
Feature weights = getWeights();
for (int b=0; b<feat.Shape().nBands; b++){
CFloatImage currentBandWeights = CFloatImage(weights.Shape().width, weights.Shape().height, 1);
CFloatImage currentBandFeatures = CFloatImage(feat.Shape().width, feat.Shape().height, 1);
CFloatImage convolved = CFloatImage(CShape(feat.Shape().width, feat.Shape().height, 1));
CFloatImage final(CShape(feat.Shape().width, feat.Shape().height, 1));
BandSelect(weights, currentBandWeights, b, 0);
BandSelect(feat, currentBandFeatures, b, 0);
currentBandWeights.origin[0] = weights.origin[0];
currentBandWeights.origin[1] = weights.origin[1];
Convolve(feat, convolved, currentBandWeights);
BandSelect(convolved, final, b, 0);
try{
score += final;
} catch (CError err) {
printf("OH NOES: the final chapter!");
}
}
score-=getBiasTerm();
/******** END TODO ********/
return score;
}
HOGFeatureExtractor::HOGFeatureExtractor(int nAngularBins, bool unsignedGradients, int cellSize):
_nAngularBins(nAngularBins),
_unsignedGradients(unsignedGradients),
_cellSize(cellSize)
{
_kernelDx.ReAllocate(CShape(3, 1, 1), derivKvals, false, 1);
_kernelDx.origin[0] = 1;
_kernelDy.ReAllocate(CShape(1, 3, 1), derivKvals, false, 1);
_kernelDy.origin[0] = 1;
// For visualization
// A set of patches representing the bin orientations. When drawing a hog cell
// we multiply each patch by the hog bin value and add all contributions up to
// form the visual representation of one cell. Full HOG is achieved by stacking
// the viz for individual cells horizontally and vertically.
_oriMarkers.resize(_nAngularBins);
const int ms = 11;
CShape markerShape(ms, ms, 1);
// First patch is a horizontal line
_oriMarkers[0].ReAllocate(markerShape, true);
_oriMarkers[0].ClearPixels();
for(int i = 1; i < ms - 1; i++) _oriMarkers[0].Pixel(/*floor(*/ ms/2 /*)*/, i, 0) = 1;
#if 0 // debug
std::cout << "DEBUG:" << __FILE__ << ":" << __LINE__ << std::endl;
for(int i = 0; i < ms; i++) {
for(int j = 0; j < ms; j++) {
std::cout << _oriMarkers[0].Pixel(j, i, 0) << " ";
}
std::cout << std::endl;
}
std::cout << std::endl;
char debugFName[2000];
sprintf(debugFName, "/tmp/debug%03d.tga", 0);
PRINT_EXPR(debugFName);
WriteFile(_oriMarkers[0], debugFName);
#endif
// The other patches are obtained by rotating the first one
CTransform3x3 T = CTransform3x3::Translation((ms - 1) / 2.0, (ms - 1) / 2.0);
for(int angBin = 1; angBin < _nAngularBins; angBin++) {
double theta;
if(unsignedGradients) theta = 180.0 * (double(angBin) / _nAngularBins);
else theta = 360.0 * (double(angBin) / _nAngularBins);
CTransform3x3 R = T * CTransform3x3::Rotation(theta) * T.Inverse();
_oriMarkers[angBin].ReAllocate(markerShape, true);
_oriMarkers[angBin].ClearPixels();
WarpGlobal(_oriMarkers[0], _oriMarkers[angBin], R, eWarpInterpLinear);
#if 0 // debug
char debugFName[2000];
sprintf(debugFName, "/tmp/debug%03d.tga", angBin);
PRINT_EXPR(debugFName);
WriteFile(_oriMarkers[angBin], debugFName);
#endif
}
}
/******************* TO DO 5 *********************
* BlendImages:
* INPUT:
* ipv: list of input images and their relative positions in the mosaic
* blendWidth: width of the blending function
* OUTPUT:
* create & return final mosaic by blending all images
* and correcting for any vertical drift
*/
CByteImage BlendImages(CImagePositionV& ipv, float blendWidth)
{
// Assume all the images are of the same shape (for now)
CByteImage& img0 = ipv[0].img;
CShape sh = img0.Shape();
int width = sh.width;
int height = sh.height;
int nBands = sh.nBands;
// int dim[2] = {width, height};
int n = ipv.size();
if (n == 0) return CByteImage(0,0,1);
bool is360 = false;
// Hack to detect if this is a 360 panorama
if (ipv[0].imgName == ipv[n-1].imgName)
is360 = true;
// Compute the bounding box for the mosaic
float min_x = FLT_MAX, min_y = FLT_MAX;
float max_x = 0, max_y = 0;
int i;
for (i = 0; i < n; i++)
{
CTransform3x3 &T = ipv[i].position;
// BEGIN TODO
// add some code here to update min_x, ..., max_y
printf("TODO: %s:%d\n", __FILE__, __LINE__);
// END TODO
}
// Create a floating point accumulation image
CShape mShape((int)(ceil(max_x) - floor(min_x)),
(int)(ceil(max_y) - floor(min_y)), nBands + 1);
CFloatImage accumulator(mShape);
accumulator.ClearPixels();
double x_init, x_final;
double y_init, y_final;
// Add in all of the images
for (i = 0; i < n; i++) {
// Compute the sub-image involved
CTransform3x3 &M = ipv[i].position;
CTransform3x3 M_t = CTransform3x3::Translation(-min_x, -min_y) * M;
CByteImage& img = ipv[i].img;
// Perform the accumulation
AccumulateBlend(img, accumulator, M_t, blendWidth);
if (i == 0) {
CVector3 p;
p[0] = 0.5 * width;
p[1] = 0.0;
p[2] = 1.0;
p = M_t * p;
x_init = p[0];
y_init = p[1];
} else if (i == n - 1) {
CVector3 p;
p[0] = 0.5 * width;
p[1] = 0.0;
p[2] = 1.0;
p = M_t * p;
x_final = p[0];
y_final = p[1];
}
}
// Normalize the results
mShape = CShape((int)(ceil(max_x) - floor(min_x)),
(int)(ceil(max_y) - floor(min_y)), nBands);
CByteImage compImage(mShape);
NormalizeBlend(accumulator, compImage);
bool debug_comp = false;
if (debug_comp)
WriteFile(compImage, "tmp_comp.tga");
// Allocate the final image shape
int outputWidth = 0;
if (is360) {
outputWidth = mShape.width - width;
} else {
outputWidth = mShape.width;
}
CShape cShape(outputWidth, mShape.height, nBands);
CByteImage croppedImage(cShape);
// Compute the affine transformation
CTransform3x3 A = CTransform3x3(); // identify transform to initialize
// BEGIN TODO
// fill in appropriate entries in A to trim the left edge and
// to take out the vertical drift if this is a 360 panorama
// (i.e. is360 is true)
printf("TODO: %s:%d\n", __FILE__, __LINE__);
// END TODO
// Warp and crop the composite
WarpGlobal(compImage, croppedImage, A, eWarpInterpLinear);
return croppedImage;
}
