BaselineData BHFitsImageSet::loadData(const ImageSetIndex &index)
{
const BHFitsImageSetIndex &fitsIndex = static_cast<const BHFitsImageSetIndex&>(index);
TimeFrequencyMetaDataPtr metaData(new TimeFrequencyMetaData());
TimeFrequencyData data;
loadImageData(data, metaData, fitsIndex);
return BaselineData(data, metaData, index);
}
////////////////////////////////////////////////////////////////////////////////
//! Run a simple benchmark test for CUDA
////////////////////////////////////////////////////////////////////////////////
void
runBenchmark( int argc, char **argv )
{
int devID = 0;
shrLog("[runBenchmark]: [%s]\n", sSDKsample);
devID = cutilChooseCudaDevice(argc, argv);
loadImageData(argc, argv);
initCuda();
g_CheckRender = new CheckBackBuffer(width, height, 4, false);
g_CheckRender->setExecPath(argv[0]);
unsigned int *d_result;
cutilSafeCall( cudaMalloc( (void **)&d_result, width*height*sizeof(unsigned int)) );
// warm-up
boxFilterRGBA(d_img, d_temp, d_temp, width, height, filter_radius, iterations, nthreads);
cutilSafeCall( cutilDeviceSynchronize() );
// Start round-trip timer and process iCycles loops on the GPU
iterations = 1; // standard 1-pass filtering
const int iCycles = 150;
double dProcessingTime = 0.0;
shrLog("\nRunning BoxFilterGPU for %d cycles...\n\n", iCycles);
shrDeltaT(2);
for (int i = 0; i < iCycles; i++)
{
dProcessingTime += boxFilterRGBA(d_img, d_temp, d_img, width, height, filter_radius, iterations, nthreads);
}
// check if kernel execution generated an error and sync host
cutilCheckMsg("Error: boxFilterRGBA Kernel execution FAILED");
cutilSafeCall(cutilDeviceSynchronize());
// Get average computation time
dProcessingTime /= (double)iCycles;
// log testname, throughput, timing and config info to sample and master logs
shrLogEx(LOGBOTH | MASTER, 0, "boxFilter-texture, Throughput = %.4f M RGBA Pixels/s, Time = %.5f s, Size = %u RGBA Pixels, NumDevsUsed = %u, Workgroup = %u\n",
(1.0e-6 * width * height)/dProcessingTime, dProcessingTime,
(width * height), 1, nthreads);
shrLog("\n");
}
////////////////////////////////////////////////////////////////////////////////
//! Run a simple benchmark test for CUDA
////////////////////////////////////////////////////////////////////////////////
int runBenchmark(int argc, char **argv)
{
printf("[runBenchmark]: [%s]\n", sSDKsample);
loadImageData(argc, argv);
initCuda();
unsigned int *dResult;
size_t pitch;
checkCudaErrors(cudaMallocPitch((void **)&dResult, &pitch, width*sizeof(unsigned int), height));
sdkStartTimer(&kernel_timer);
// warm-up
bilateralFilterRGBA(dResult, width, height, euclidean_delta, filter_radius, iterations, kernel_timer);
checkCudaErrors(cudaDeviceSynchronize());
// Start round-trip timer and process iCycles loops on the GPU
iterations = 1; // standard 1-pass filtering
const int iCycles = 150;
double dProcessingTime = 0.0;
printf("\nRunning BilateralFilterGPU for %d cycles...\n\n", iCycles);
for (int i = 0; i < iCycles; i++)
{
dProcessingTime += bilateralFilterRGBA(dResult, width, height, euclidean_delta, filter_radius, iterations, kernel_timer);
}
// check if kernel execution generated an error and sync host
getLastCudaError("Error: bilateralFilterRGBA Kernel execution FAILED");
checkCudaErrors(cudaDeviceSynchronize());
sdkStopTimer(&kernel_timer);
// Get average computation time
dProcessingTime /= (double)iCycles;
// log testname, throughput, timing and config info to sample and master logs
printf("bilateralFilter-texture, Throughput = %.4f M RGBA Pixels/s, Time = %.5f s, Size = %u RGBA Pixels, NumDevsUsed = %u\n",
(1.0e-6 * width * height)/dProcessingTime, dProcessingTime, (width * height), 1);
printf("\n");
return 0;
}
void runAutoTest(int argc, char **argv, const char *dump_filename, eFilterMode filter_mode)
{
cudaDeviceProp deviceProps;
int devID = findCudaDevice(argc, (const char **)argv);
checkCudaErrors(cudaGetDeviceProperties(&deviceProps, devID));
printf("[%s] (automated testing w/ readback)\n", sSDKsample);
printf("CUDA device [%s] has %d Multi-Processors\n", deviceProps.name, deviceProps.multiProcessorCount);
loadImageData(argc, argv);
uchar4 *d_output;
checkCudaErrors(cudaMalloc((void **)&d_output, imageWidth*imageHeight*4));
unsigned int *h_result = (unsigned int *)malloc(width * height * sizeof(unsigned int));
printf("AutoTest: %s Filter Mode: <%s>\n", sSDKsample, sFilterMode[g_FilterMode]);
render(imageWidth, imageHeight,
tx, ty, scale, cx, cy,
blockSize, gridSize, filter_mode, d_output);
// check if kernel execution generated an error
getLastCudaError("Error: render (bicubicTexture) Kernel execution FAILED");
checkCudaErrors(cudaDeviceSynchronize());
cudaMemcpy(h_result, d_output, imageWidth*imageHeight*4, cudaMemcpyDeviceToHost);
sdkSavePPM4ub(dump_filename, (unsigned char *)h_result, imageWidth, imageHeight);
checkCudaErrors(cudaFree(d_output));
free(h_result);
// cudaDeviceReset causes the driver to clean up all state. While
// not mandatory in normal operation, it is good practice. It is also
// needed to ensure correct operation when the application is being
// profiled. Calling cudaDeviceReset causes all profile data to be
// flushed before the application exits
cudaDeviceReset();
}
void runAutoTest(int argc, char **argv)
{
int devID = 0;
shrLog("[runAutoTest]: [%s] (automated testing w/ readback)\n", sSDKsample);
devID = cutilChooseCudaDevice(argc, argv);
loadImageData(argc, argv);
initCuda();
g_CheckRender = new CheckBackBuffer(width, height, 4, false);
g_CheckRender->setExecPath(argv[0]);
unsigned int *d_result;
cutilSafeCall( cudaMalloc( (void **)&d_result, width*height*sizeof(unsigned int)) );
for(int i = 0; i < 4; i++)
{
shrLog("[AutoTest]: %s (radius=%d)", sSDKsample, filter_radius );
bilateralFilterRGBA(d_result, width, height, euclidean_delta, filter_radius, iterations, nthreads);
// check if kernel execution generated an error
cutilCheckMsg("Error: bilateralFilterRGBA Kernel execution FAILED");
cutilSafeCall( cutilDeviceSynchronize() );
cudaMemcpy(g_CheckRender->imageData(), d_result, width*height*sizeof(unsigned int), cudaMemcpyDeviceToHost);
g_CheckRender->savePPM(sOriginal[i], false, NULL);
if (!g_CheckRender->PPMvsPPM(sOriginal[i], sReference[i], MAX_EPSILON_ERROR, 0.15f)) {
g_TotalErrors++;
}
gaussian_delta += 1.0f;
euclidean_delta *= 1.25f;
updateGaussian(gaussian_delta, filter_radius);
}
cutilSafeCall( cudaFree( d_result ) );
delete g_CheckRender;
}
void
ProcessImage::insertNewData(int index, QList<struct DetectorData> detectedData){
showCase->setDone( index );
if (detectedData.size() == 0){
detectedFaces->noFace();
} else {
QList<struct FaceData> facedata;
struct FaceData f;
for (int i = 0; i < detectedData.size(); i++ ){
f.rect = detectedData.at(i).rect;
f.image = detectedData.at(i).image;
f.personId = -1;
f.eigen_ratio = 0;
f.fisher_ratio = 0;
f.lbph_ratio = 0;
f.voted = false;
facedata << f;
}
showCase->addCoordinates( index, facedata );
}
// RECOGNIZE THE FACES
// recognizer = new Recognizer();
for (int i = 0; i < detectedData.size(); i++ ){
qDebug() << "send image " << i << " to recognition";
Socket::sock().recognize( detectedData.at(i).mat.clone(), index, i );
// int personId = recognizer->recognize( detectedData.at(i) );
// showCase->setPersonId(index, i, personId); // image index, face index, person id
}
/* show if current */
if (index == _currentImage){
loadImageData( index );
}
}
void initialize(int argc, char **argv)
{
printf("[%s] (OpenGL Mode)\n", sSDKsample);
// First initialize OpenGL context, so we can properly set the GL for CUDA.
// This is necessary in order to achieve optimal performance with OpenGL/CUDA interop.
initGL(&argc, argv);
int devID;
cudaDeviceProp deviceProps;
if (checkCmdLineFlag(argc, (const char **)argv, "device"))
{
devID = gpuGLDeviceInit(argc, (const char **)argv);
if (devID < 0)
{
printf("exiting...\n");
exit(EXIT_SUCCESS);
}
}
else
{
devID = gpuGetMaxGflopsDeviceId();
cudaGLSetGLDevice(devID);
}
// get number of SMs on this GPU
checkCudaErrors(cudaGetDeviceProperties(&deviceProps, devID));
printf("CUDA device [%s] has %d Multi-Processors\n", deviceProps.name, deviceProps.multiProcessorCount);
// Create the timer (for fps measurement)
sdkCreateTimer(&timer);
// load image from disk
loadImageData(argc, argv);
printf("\n"
"\tControls\n"
"\t=/- : Zoom in/out\n"
"\tb : Run Benchmark g_FilterMode\n"
"\tc : Draw Bicubic Spline Curve\n"
"\t[esc] - Quit\n\n"
"\tPress number keys to change filtering g_FilterMode:\n\n"
"\t1 : nearest filtering\n"
"\t2 : bilinear filtering\n"
"\t3 : bicubic filtering\n"
"\t4 : fast bicubic filtering\n"
"\t5 : Catmull-Rom filtering\n\n"
);
initGLBuffers();
#if USE_BUFFER_TEX
fprog = compileASMShader(GL_FRAGMENT_PROGRAM_ARB, shaderCode);
if (!fprog)
{
exit(EXIT_SUCCESS);
}
#endif
}
////////////////////////////////////////////////////////////////////////////////
// Program main
////////////////////////////////////////////////////////////////////////////////
int main(int argc, char **argv)
{
// start logs
int devID;
char *ref_file = NULL;
printf("%s Starting...\n\n", argv[0]);
#if defined(__linux__)
setenv ("DISPLAY", ":0", 0);
#endif
// use command-line specified CUDA device, otherwise use device with highest Gflops/s
if (argc > 1)
{
if (checkCmdLineFlag(argc, (const char **)argv, "radius"))
{
filter_radius = getCmdLineArgumentInt(argc, (const char **) argv, "radius");
}
if (checkCmdLineFlag(argc, (const char **)argv, "passes"))
{
iterations = getCmdLineArgumentInt(argc, (const char **)argv, "passes");
}
if (checkCmdLineFlag(argc, (const char **)argv, "file"))
{
getCmdLineArgumentString(argc, (const char **)argv, "file", (char **)&ref_file);
}
}
// load image to process
loadImageData(argc, argv);
if (checkCmdLineFlag(argc, (const char **)argv, "benchmark"))
{
// This is a separate mode of the sample, where we are benchmark the kernels for performance
devID = findCudaDevice(argc, (const char **)argv);
// Running CUDA kernels (bilateralfilter) in Benchmarking mode
g_TotalErrors += runBenchmark(argc, argv);
// cudaDeviceReset causes the driver to clean up all state. While
// not mandatory in normal operation, it is good practice. It is also
// needed to ensure correct operation when the application is being
// profiled. Calling cudaDeviceReset causes all profile data to be
// flushed before the application exits
cudaDeviceReset();
exit(g_TotalErrors == 0 ? EXIT_SUCCESS : EXIT_FAILURE);
}
else if (checkCmdLineFlag(argc, (const char **)argv, "radius") ||
checkCmdLineFlag(argc, (const char **)argv, "passes"))
{
// This overrides the default mode. Users can specify the radius used by the filter kernel
devID = findCudaDevice(argc, (const char **)argv);
g_TotalErrors += runSingleTest(ref_file, argv[0]);
// cudaDeviceReset causes the driver to clean up all state. While
// not mandatory in normal operation, it is good practice. It is also
// needed to ensure correct operation when the application is being
// profiled. Calling cudaDeviceReset causes all profile data to be
// flushed before the application exits
cudaDeviceReset();
exit(g_TotalErrors == 0 ? EXIT_SUCCESS : EXIT_FAILURE);
}
else
{
// Default mode running with OpenGL visualization and in automatic mode
// the output automatically changes animation
printf("\n");
// First initialize OpenGL context, so we can properly set the GL for CUDA.
// This is necessary in order to achieve optimal performance with OpenGL/CUDA interop.
initGL(argc, (char **)argv);
int dev = findCapableDevice(argc, argv);
if (dev != -1)
{
dev = gpuGLDeviceInit(argc, (const char **)argv);
if (dev == -1)
{
exit(EXIT_FAILURE);
}
}
else
{
// cudaDeviceReset causes the driver to clean up all state. While
// not mandatory in normal operation, it is good practice. It is also
// needed to ensure correct operation when the application is being
// profiled. Calling cudaDeviceReset causes all profile data to be
// flushed before the application exits
cudaDeviceReset();
exit(EXIT_SUCCESS);
}
// Now we can create a CUDA context and bind it to the OpenGL context
initCuda();
initGLResources();
// sets the callback function so it will call cleanup upon exit
#if defined (__APPLE__) || defined(MACOSX)
atexit(cleanup);
#else
glutCloseFunc(cleanup);
#endif
printf("Running Standard Demonstration with GLUT loop...\n\n");
printf("Press '+' and '-' to change filter width\n"
"Press ']' and '[' to change number of iterations\n"
"Press 'e' and 'E' to change Euclidean delta\n"
"Press 'g' and 'G' to changle Gaussian delta\n"
"Press 'a' or 'A' to change Animation mode ON/OFF\n\n");
// Main OpenGL loop that will run visualization for every vsync
glutMainLoop();
}
}
////////////////////////////////////////////////////////////////////////////////
// Program main
////////////////////////////////////////////////////////////////////////////////
int
main( int argc, char** argv)
{
pArgc = &argc;
pArgv = argv;
shrQAStart(argc, argv);
// start logs
shrSetLogFileName ("boxFilter.txt");
shrLog("%s Starting...\n\n", argv[0]);
// use command-line specified CUDA device, otherwise use device with highest Gflops/s
if (argc > 1) {
cutGetCmdLineArgumenti( argc, (const char**) argv, "threads", &nthreads );
cutGetCmdLineArgumenti( argc, (const char**) argv, "radius", &filter_radius);
if (cutCheckCmdLineFlag(argc, (const char **)argv, "glverify"))
{
g_bOpenGLQA = true;
fpsLimit = frameCheckNumber;
}
if (cutCheckCmdLineFlag(argc, (const char **)argv, "fbo")) {
g_bFBODisplay = true;
}
}
// load image to process
loadImageData(argc, argv);
if (cutCheckCmdLineFlag(argc, (const char **)argv, "qatest") ||
cutCheckCmdLineFlag(argc, (const char **)argv, "noprompt"))
{
// Running CUDA kernel (boxFilter) without visualization (QA Testing/Verification)
runAutoTest(argc, argv);
shrQAFinishExit(argc, (const char **)argv, (g_TotalErrors == 0 ? QA_PASSED : QA_FAILED));
}
else if (cutCheckCmdLineFlag(argc, (const char **)argv, "benchmark"))
{
// Running CUDA kernels (boxfilter) in Benchmarking mode
runBenchmark(argc, argv);
shrQAFinishExit(argc, (const char **)argv, (g_TotalErrors == 0 ? QA_PASSED : QA_FAILED));
}
else
{
// Running CUDA kernels (boxFilter) with OpenGL visualization
if (g_bFBODisplay) shrLog("[FBO Display] ");
if (g_bOpenGLQA) shrLog("[OpenGL Readback Comparisons] ");
shrLog("\n");
if ( cutCheckCmdLineFlag(argc, (const char **)argv, "device")) {
printf(" This SDK does not explicitly support -device=n when running with OpenGL.\n");
printf(" When specifying -device=n (n=0,1,2,....) the sample must not use OpenGL.\n");
printf(" See details below to run without OpenGL:\n\n");
printf(" > %s -device=n -qatest\n\n", argv[0]);
printf("exiting...\n");
shrQAFinishExit(argc, (const char **)argv, QA_WAIVED);
}
// First initialize OpenGL context, so we can properly set the GL for CUDA.
// This is necessary in order to achieve optimal performance with OpenGL/CUDA interop.
initGL( &argc, argv );
int dev = findCapableDevice(argc, argv);
if( dev != -1 ) {
cudaGLSetGLDevice( dev );
} else {
cutilDeviceReset();
shrQAFinishExit(argc, (const char **)argv, QA_WAIVED);
}
// Now we can create a CUDA context and bind it to the OpenGL context
initCuda();
initGLResources();
if (g_bOpenGLQA) {
if (g_bFBODisplay) {
g_CheckRender = new CheckFBO(width, height, 4, g_FrameBufferObject);
} else {
g_CheckRender = new CheckBackBuffer(width, height, 4);
}
g_CheckRender->setPixelFormat(GL_RGBA);
g_CheckRender->setExecPath(argv[0]);
g_CheckRender->EnableQAReadback(true);
}
}
// sets the callback function so it will call cleanup upon exit
atexit(cleanup);
shrLog("Running Standard Demonstration with GLUT loop...\n\n");
shrLog("Press '+' and '-' to change filter width\n"
"Press ']' and '[' to change number of iterations\n\n");
// Main OpenGL loop that will run visualization for every vsync
glutMainLoop();
cutilDeviceReset();
shrQAFinishExit(argc, (const char **)argv, (g_TotalErrors == 0 ? QA_PASSED : QA_FAILED));
}
////////////////////////////////////////////////////////////////////////////////
// Program main
////////////////////////////////////////////////////////////////////////////////
int main( int argc, char** argv)
{
shrQAStart(argc, argv);
// start logs
shrSetLogFileName ("bilateralFilter.txt");
shrLog("%s Starting...\n\n", argv[0]);
// use command-line specified CUDA device, otherwise use device with highest Gflops/s
cutGetCmdLineArgumenti( argc, (const char**) argv, "threads", &nthreads );
cutGetCmdLineArgumenti( argc, (const char**) argv, "radius", &filter_radius);
// load image to process
loadImageData(argc, argv);
if (cutCheckCmdLineFlag(argc, (const char **)argv, "qatest") ||
cutCheckCmdLineFlag(argc, (const char **)argv, "noprompt"))
{
// Running CUDA kernel (bilateralFilter) without visualization (QA Testing/Verification)
runAutoTest(argc, argv);
shrQAFinishExit(argc, (const char **)argv, (g_TotalErrors == 0 ? QA_PASSED : QA_FAILED));
}
else if (cutCheckCmdLineFlag(argc, (const char **)argv, "benchmark"))
{
// Running CUDA kernel (bilateralFilter) in Benchmarking Mode
runBenchmark(argc, argv);
shrQAFinishExit(argc, (const char **)argv, (g_TotalErrors == 0 ? QA_PASSED : QA_FAILED));
}
else
{
// Running CUDA kernel (bilateralFilter) in CUDA + OpenGL Visualization Mode
if ( cutCheckCmdLineFlag(argc, (const char **)argv, "device")) {
printf("[%s]\n", argv[0]);
printf(" Does not explicitly support -device=n in OpenGL mode\n");
printf(" To use -device=n, the sample must be running w/o OpenGL\n\n");
printf(" > %s -device=n -qatest\n", argv[0]);
printf("exiting...\n");
exit(0);
}
// First initialize OpenGL context, so we can properly set the GL for CUDA.
// This is necessary in order to achieve optimal performance with OpenGL/CUDA interop.
initGL( argc, argv );
if ( cutCheckCmdLineFlag(argc, (const char **)argv, "device")) {
cutilGLDeviceInit(argc, argv);
} else {
cudaGLSetGLDevice (cutGetMaxGflopsDeviceId() );
}
initCuda();
initOpenGL();
}
atexit(cleanup);
printf("Running Standard Demonstration with GLUT loop...\n\n");
printf("Press '+' and '-' to change number of iterations\n"
"Press LEFT and RIGHT change euclidean delta\n"
"Press UP and DOWN to change gaussian delta\n"
"Press '1' to show original image\n"
"Press '2' to show result\n\n");
glutMainLoop();
cutilDeviceReset();
shrQAFinishExit(argc, (const char **)argv, (g_TotalErrors == 0 ? QA_PASSED : QA_FAILED));
}
ImageView::ImageView(QWidget * parent) :
QGraphicsView(parent)
{
loadImageData(ImageView::sample1src);
}
/**
* Reloads original image
*/
void ImageView::reloadImageData()
{
loadImageData(imageSource);
}
////////////////////////////////////////////////////////////////////////////////
// Program main
////////////////////////////////////////////////////////////////////////////////
int
main(int argc, char **argv)
{
int devID = 0;
char *ref_file = NULL;
pArgc = &argc;
pArgv = argv;
// start logs
printf("%s Starting...\n\n", argv[0]);
if (checkCmdLineFlag(argc, (const char **)argv, "help"))
{
printHelp();
exit(EXIT_SUCCESS);
}
// use command-line specified CUDA device, otherwise use device with highest Gflops/s
if (argc > 1)
{
if (checkCmdLineFlag(argc, (const char **)argv, "threads"))
{
nthreads = getCmdLineArgumentInt(argc, (const char **) argv, "threads");
}
if (checkCmdLineFlag(argc, (const char **)argv, "radius"))
{
filter_radius = getCmdLineArgumentInt(argc, (const char **) argv, "radius");
}
if (checkCmdLineFlag(argc, (const char **)argv, "passes"))
{
iterations = getCmdLineArgumentInt(argc, (const char **) argv, "passes");
}
if (checkCmdLineFlag(argc, (const char **)argv, "file"))
{
getCmdLineArgumentString(argc, (const char **)argv, "file", (char **)&ref_file);
}
}
// load image to process
loadImageData(argc, argv);
if (checkCmdLineFlag(argc, (const char **)argv, "benchmark"))
{
// This is a separate mode of the sample, where we are benchmark the kernels for performance
devID = findCudaDevice(argc, (const char **)argv);
// Running CUDA kernels (boxfilter) in Benchmarking mode
g_TotalErrors += runBenchmark();
exit(g_TotalErrors == 0 ? EXIT_SUCCESS : EXIT_FAILURE);
}
else if (checkCmdLineFlag(argc, (const char **)argv, "radius") ||
checkCmdLineFlag(argc, (const char **)argv, "passes"))
{
// This overrides the default mode. Users can specify the radius used by the filter kernel
devID = findCudaDevice(argc, (const char **)argv);
g_TotalErrors += runSingleTest(ref_file, argv[0]);
exit(g_TotalErrors == 0 ? EXIT_SUCCESS : EXIT_FAILURE);
}
else
{
// Default mode running with OpenGL visualization and in automatic mode
// the output automatically changes animation
printf("\n");
// First initialize OpenGL context, so we can properly set the GL for CUDA.
// This is necessary in order to achieve optimal performance with OpenGL/CUDA interop.
initGL(&argc, argv);
int dev = findCapableDevice(argc, argv);
if (dev != -1)
{
cudaGLSetGLDevice(dev);
}
else
{
cudaDeviceReset();
exit(EXIT_SUCCESS);
}
// Now we can create a CUDA context and bind it to the OpenGL context
initCuda();
initGLResources();
// sets the callback function so it will call cleanup upon exit
atexit(cleanup);
printf("Running Standard Demonstration with GLUT loop...\n\n");
printf("Press '+' and '-' to change filter width\n"
"Press ']' and '[' to change number of iterations\n"
"Press 'a' or 'A' to change animation ON/OFF\n\n");
// Main OpenGL loop that will run visualization for every vsync
glutMainLoop();
cudaDeviceReset();
exit(g_TotalErrors == 0 ? EXIT_SUCCESS : EXIT_FAILURE);
}
}
