TEST_F(for_each_channel_accumulate_test, should_throw_when_image_dimensions_dont_match)
{
rgb16_image_t img1(1, 2), img2(1, 3), img3(2, 2);
auto zero = [](channel16_t, channel16_t) { return 0; };
ASSERT_THROW(for_each_channel_accumulate(const_view(img1), const_view(img2), 0, zero), std::invalid_argument);
ASSERT_THROW(for_each_channel_accumulate(const_view(img1), const_view(img3), 0, zero), std::invalid_argument);
}
TEST_F(for_each_channel_accumulate_test, should_return_initial_value_for_empty_images)
{
rgb16_image_t img1, img2;
ASSERT_EQ(8, for_each_channel_accumulate(const_view(img1), const_view(img2), 8, [&](channel16_t p1, channel16_t p2)
{
return 10;
}));
}
void FrameAverager::addFrame(const RawImage& frame)
{
if (view(accumulator).size() == 0)
{
accumulator.recreate(frame.dimensions());
fill_pixels(view(accumulator), Color::black());
}
boost::gil::transform_pixels(const_view(frame), const_view(accumulator), view(accumulator), boost::gil::pixel_plus_t<RawPixel, AccumPixel, AccumPixel>());
++frameCount;
}
void ImageGetter::resizeJPGImage(std::string const& imageURL, int sixex, int sizey)
{
boost::gil::rgb8_image_t img;
jpeg_read_image(imageURL,img);
boost::gil::rgb8_image_t targetImageSize(sixex,sizey);
resize_view(const_view(img), view(targetImageSize), boost::gil::bilinear_sampler());
jpeg_write_view(imageURL,const_view(targetImageSize));
}
TEST_F(for_each_channel_accumulate_test, should_include_the_initial_value)
{
rgb16_image_t img1(1, 1), img2(1, 1);
view(img1)(0, 0) = { 1, 0, 0 };
view(img2)(0, 0) = { 1, 0, 0 };
ASSERT_EQ(2 + 8, for_each_channel_accumulate(const_view(img1), const_view(img2), 8, [&](channel16_t p1, channel16_t p2)
{
return p1 + p2;
}));
}
TEST_F(for_each_channel_accumulate_test, should_call_functor_for_each_channel_in_parallel_in_both_images_and_sum_the_results)
{
rgb16_image_t img1(1, 1);
rgb8_image_t img2(1, 1);
view(img1)(0, 0) = { 1025, 18, 36 };
view(img2)(0, 0) = { 1, 2, 4 };
ASSERT_EQ(1024 + 16 + 32, for_each_channel_accumulate(const_view(img1), const_view(img2), 0, [&](channel16_t p1, channel8_t p2)
{
return p1 - p2;
})) << "should sum all results";
}
TEST_F(for_each_channel_accumulate_test, should_call_functor_for_each_pixel_in_parallel_in_both_images)
{
rgb16_image_t img1(3, 2), img2(3, 2);
auto v1 = view(img1), v2 = view(img2);
v1(0, 0) = { 1, 0, 0 }; v1(1, 0) = { 2, 0, 0 }; v1(2, 0) = { 4, 0, 0 };
v1(0, 1) = { 8, 0, 0 }; v1(1, 1) = { 16, 0, 0 }; v1(2, 1) = { 32, 0, 0 };
v2(0, 0) = { 64, 0, 0 }; v2(1, 0) = { 128, 0, 0 }; v2(2, 0) = { 256, 0, 0 };
v2(0, 1) = { 512, 0, 0 }; v2(1, 1) = { 1024, 0, 0 }; v2(2, 1) = { 2048, 0, 0 };
ASSERT_EQ(4095, for_each_channel_accumulate(const_view(img1), const_view(img2), 0, [&](channel16_t p1, channel16_t p2)
{
return p1 + p2;
})) << "should call functor for each pixel";
}
void save_image(
cell cells[GRID_W][GRID_H],
Vec2 particles[PARTICLES_NUM],
std::string filename) {
int w = GRID_W;
int h = GRID_H;
rgb8_image_t img(w, h);
rgb8_image_t::view_t v = view(img);
for(int i = 0; i < w; i++){
for(int j = 0; j < h; j++){
cell * cell = &cells[i][j];
Vec2 * cu = &cell->u;
v(i,j) = rgb8_pixel_t(cu->x * 255 + 128, cu->y * 255 + 128, 0);
//v(i,j) = rgb8_pixel_t(0, 0, 0);
}
}
for(int i = 0; i < PARTICLES_NUM; i++){
int posi = particles[i].x;
int posj = particles[i].y;
if(posi < 0 || posi >= w || posj < 0 || posj >= h){
continue;
}
v(posi, posj) = rgb8_pixel_t(255, 255, 255);
}
png_write_view(filename, const_view(img));
}
void rgba_gaussian_pyramid_t::build()
{
float sigma = ( 1.0f / ratio_ - 1.0f);
buffer_t tmp( const_view().width(), const_view().height(), 4);
buffer_t tmp2( tmp.height(), tmp.width(), 4);
for( int i = 1; i < num_levels_; ++i)
{
int w = const_view( i-1).width();
int h = const_view( i-1).height();
image::image_view_t tmp_view( boost::gil::subimage_view( tmp.rgba_view(), 0, 0, w, h));
gaussian_blur_rgba( const_view( i-1), boost::gil::subimage_view( tmp2.rgba_view(), 0, 0, h, w), tmp_view, sigma, sigma);
scale_view( tmp_view, view( i));
}
}
inline void PrintTo(const Nebula::RawImage& img, ::std::ostream* os)
{
*os << "RawImage[ " << img.width() << " x " << img.height() << " :";
for (auto p : const_view(img))
{
*os << " (";
static_for_each(p, [=](Nebula::RawChannel ch) { *os << " " << ch; });
*os << " )";
}
*os << " ]";
}
int FeaturesDetectionApplication::main_loop(program_options::variables_map &options)
{
printf("FeaturesDetectionApplication::main_loop says hello world !\n");
//init_gui(options);
//run_gui();
// initialization ---
gst_video_input_p.reset(new GstVideoInput(options));
features_detector_p.reset(new SimpleFAST(options));
// video output ---
rgb8_cimg_t current_image(gst_video_input_p->get_image_dimensions());
gst_video_input_p->get_new_image(current_image.view); // copy the data
CImgDisplay video_display(current_image.dimx(), current_image.dimy(), get_application_title().c_str());
video_display.show();
video_display.display(current_image);
// intermediary image --
gray8_image_t gray_image(current_image.view.dimensions());
// main loop ---
do
{
// get new image --
gst_video_input_p->get_new_image(current_image.view); // copy the data
// color to gray_image
copy_and_convert_pixels(current_image.view, boost::gil::view(gray_image));
// compute features
const vector<FASTFeature> &features =
features_detector_p->detect_features((const_view(gray_image)));
// plot features on output image
draw_features(features, current_image);
video_display.display(current_image);
// add a delay ---
wait_some_seconds(0.1); // [seconds]
}
while (video_display.is_closed == false);
return 0;
}
void boost_parse_image(std::string filename, size_t& width, size_t& height, size_t& channels, Format& format, size_t& image_data_size, std::string format_string) {
typedef boost::mpl::vector<gray8_image_t, gray16_image_t, rgb8_image_t, rgb16_image_t> my_img_types;
any_image<my_img_types> src_image;
if (format_string == "JPG"){
jpeg_read_image(filename, src_image);
format = Format::JPG;
} else if (format_string == "PNG"){
png_read_image(filename, src_image);
format = Format::PNG;
} else{
if (boost::algorithm::ends_with(filename, "jpg") ||
boost::algorithm::ends_with(filename, "jpeg")) {
jpeg_read_image(filename, src_image);
format = Format::JPG;
} else if (boost::algorithm::ends_with(filename, "png")){
png_read_image(filename, src_image);
format = Format::PNG;
} else {
log_and_throw(std::string("Unsupported format."));
}
}
// create a view of the image
auto src_view = const_view(src_image);
// extract image information
width = src_view.width();
height = src_view.height();
channels = src_view.num_channels();
image_data_size = width*height*channels;
// Debug
// std::cout << "Read image "<< filename << "\n"
// << "width: " << width << "\n"
// << "height: " << height << "\n"
// << "num_channels " << channels << "\n"
// << std::endl;
}
///////////////////////////////////////////////////////////////////////////////////
/// 指定窓の画面をキャプチャして指定ファイルにPNG形式で保存する
void CaptureScreen(CWnd *pWnd, const std::wstring &szFilePath) {
CRect rectClient;
pWnd->GetClientRect( &rectClient );
CClientDC dc(pWnd);
CDC dc2;
dc2.CreateCompatibleDC( &dc );
CBitmap bitmap;
bitmap.CreateCompatibleBitmap(&dc, rectClient.Width(), rectClient.Height() );
auto pOldBitmap = dc2.SelectObject( &bitmap );
dc2.BitBlt(0,0, rectClient.Width(), rectClient.Height(), &dc, 0, 0, SRCCOPY );
int nWidthAdjust = rectClient.Width();
if ( nWidthAdjust % 4 ) { nWidthAdjust += 4 - nWidthAdjust % 4;}
std::vector<BYTE> bitdata( nWidthAdjust * rectClient.Height() * 3 );
dc2.SelectObject( pOldBitmap );
BITMAPINFO bitmapinfo;
ZeroMemory(&bitmapinfo, sizeof(BITMAPINFO) );
bitmapinfo.bmiHeader.biSize = sizeof(BITMAPINFOHEADER);
bitmapinfo.bmiHeader.biWidth = nWidthAdjust;
bitmapinfo.bmiHeader.biHeight = rectClient.Height();
bitmapinfo.bmiHeader.biPlanes = 1;
bitmapinfo.bmiHeader.biBitCount = 24;
bitmapinfo.bmiHeader.biCompression = BI_RGB;
GetDIBits( dc2, bitmap, 0, rectClient.Height(), bitdata.data(), &bitmapinfo, DIB_RGB_COLORS );
boost::gil::rgb8_image_t img( nWidthAdjust, rectClient.Height() );
copy_pixels(boost::gil::interleaved_view(nWidthAdjust, rectClient.Height(), (const boost::gil::rgb8_pixel_t*)bitdata.data(), nWidthAdjust*3), view(img));
png_write_view(CStringA(szFilePath.c_str()), boost::gil::flipped_up_down_view( const_view(img) ) );
}
void TiffImageWriter::writeImage(const std::string& filename, const RawImage& image)
{
boost::gil::tiff_write_view(filename, const_view(image));
}
int main(int argc, char **argv)
{
int N = 1 << 20;
// Parse custom command line args
for (int i = 1; i < argc; ++i) {
if (strcmp(argv[i],"-N") == 0) {
N = atoi(argv[++i]);
}
}
// Round up to the nearest square number
int n_side = int(std::ceil(std::sqrt(N)));
N = n_side * n_side;
// Init the FMM Kernel and options
FMMOptions opts = get_options(argc, argv);
//typedef UnitExpansion kernel_type;
//typedef ExpExpansion kernel_type;
typedef LaplaceSpherical kernel_type;
//typedef YukawaCartesian kernel_type;
// Init kernel
kernel_type K;
typedef kernel_type::point_type point_type;
typedef kernel_type::source_type source_type;
typedef kernel_type::target_type target_type;
typedef kernel_type::charge_type charge_type;
typedef kernel_type::result_type result_type;
std::cout << "Initializing source and N = " << N << " targets..." << std::endl;
// Init a square targets
double xmin = -1;
double xmax = 1;
double ymin = -1;
double ymax = 1;
std::vector<target_type> targets;
targets.reserve(N);
for (int n = 0; n < n_side; ++n) {
for (int m = 0; m < n_side; ++m) {
targets.push_back(target_type(xmin + n * (xmax-xmin) / (n_side-1),
ymin + m * (ymax-ymin) / (n_side-1)));
}
}
//int middle = n_side/2 * n_side + n_side/2;
int middle = fmmtl::random<unsigned>::get(0, N);
// Init charges, only the middle source has a charge
std::vector<charge_type> charges(N);
charges[middle] = charge_type(1);
std::cout << "Building the kernel matrix..." << std::endl;
// Build the FMM
fmmtl::kernel_matrix<kernel_type> A = K(targets, targets);
A.set_options(opts);
std::cout << "Performing the kernel matrix-vector mult..." << std::endl;
// Execute the FMM
std::vector<result_type> result = A * charges;
// Check the result
std::cout << "Computing direct kernel matrix-vector mult..." << std::endl;
// Compute the result with a direct matrix-vector multiplication
std::vector<result_type> exact(N);
fmmtl::direct(K,
targets.begin()+middle, targets.begin()+middle+1,
charges.begin()+middle,
targets.begin(), targets.end(),
exact.begin());
std::cout << "Computing the errors..." << std::endl;
std::vector<double> log_error(result.size());
double min_error = std::numeric_limits<double>::max();
double max_error = std::numeric_limits<double>::lowest();
for (unsigned k = 0; k < result.size(); ++k) {
double error = norm_2(exact[k] - result[k]) / norm_2(exact[k]);
if (error > 1e-15)
log_error[k] = std::log10(error);
else
log_error[k] = -16;
min_error = std::min(min_error, log_error[k]);
max_error = std::max(max_error, log_error[k]);
}
std::cout << "Min log error: " << min_error << std::endl;
std::cout << "Max log error: " << max_error << std::endl;
// Fill the image with the errors computed above
gil::rgb8_image_t img(n_side, n_side);
auto img_view = gil::view(img);
for (int n = 0; n < n_side; ++n) {
for (int m = 0; m < n_side; ++m) {
img_view(n,m) = make_heat((log_error[n*n_side+m] - min_error) / (max_error - min_error));
}
}
gil::png_write_view("fmmtl_errors.png", const_view(img));
}
template <typename Image> result_type operator()(const Image& img) const { return result_type(const_view(img)); }
RawImage FrameAverager::getCombinedFrame() const
{
RawImage output(accumulator.dimensions());
boost::gil::view_divides_scalar<AccumPixel>(const_view(accumulator), frameCount, view(output));
return output;
}