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;
}
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 ********/
}
