int main(int argc, char **argv) {
float Score;
float BestScore = HugeScore;
int p, i;
SetupParams(argc, argv);
clock_t Clock0;
KK K1; // main KK class, for all data
K1.penaltyMix = PenaltyMix;
Clock0 = clock(); // start timer
K1.LoadData(); // load .fet file
// Seed random number generator
srand(RandomSeed);
// open distance dump file if required
if (DistDump) Distfp = fopen("DISTDUMP", "w");
// start with provided file, if required
if (*StartCluFile) {
Output("Starting from cluster file %s\n", StartCluFile);
BestScore = K1.CEM(StartCluFile, 1, 1);
Output("%d->%d Clusters: Score %f\n\n", K1.nStartingClusters, K1.nClustersAlive, BestScore);
for(p=0; p<K1.nPoints; p++) K1.BestClass[p] = K1.Class[p];
SaveOutput(K1.BestClass);
}
// loop through numbers of clusters ...
for(K1.nStartingClusters=MinClusters; K1.nStartingClusters<=MaxClusters; K1.nStartingClusters++) for(i=0; i<nStarts; i++) {
// do CEM iteration
Output("Starting from %d clusters...\n", K1.nStartingClusters);
Score = K1.Cluster();
Output("%d->%d Clusters: Score %f, best is %f\n", K1.nStartingClusters, K1.nClustersAlive, Score, BestScore);
if (Score < BestScore) {
Output("THE BEST YET!\n");
// New best classification found
BestScore = Score;
for(p=0; p<K1.nPoints; p++) K1.BestClass[p] = K1.Class[p];
SaveOutput(K1.BestClass);
}
Output("\n");
}
SaveOutput(K1.BestClass);
Output("That took %f seconds.\n", (clock()-Clock0)/(float) CLOCKS_PER_SEC);
if (DistDump) fclose(Distfp);
return 0;
}
bool ImageSegmenter::Process()
{
// Read input image
m_docImage = imread(m_document_image_path.c_str(), CV_LOAD_IMAGE_GRAYSCALE); // converts to grayscale if required
PreprocessImage(m_docImage);
SplitIntoLines(m_docImage);
if (m_output_level == "LINES") {
SaveOutput(m_output_level);
return true;
}
else {
SplitLinesIntoWords(m_docImage);
SaveOutput(m_output_level);
}
return true;
}
/*---------------------------------------------------------------*/
int main(int argc, char *argv[]) {
int nargs, n, err;
char tmpstr[2000], *signstr=NULL,*SUBJECTS_DIR, fname[2000];
//char *OutDir = NULL;
RFS *rfs;
int nSmoothsPrev, nSmoothsDelta;
MRI *z, *zabs=NULL, *sig=NULL, *p=NULL;
int FreeMask = 0;
int nthSign, nthFWHM, nthThresh;
double sigmax, zmax, threshadj, csize, csizeavg, searchspace,avgvtxarea;
int csizen;
int nClusters, cmax,rmax,smax;
SURFCLUSTERSUM *SurfClustList;
struct timeb mytimer;
LABEL *clabel;
FILE *fp, *fpLog=NULL;
nargs = handle_version_option (argc, argv, vcid, "$Name: stable5 $");
if (nargs && argc - nargs == 1) exit (0);
argc -= nargs;
cmdline = argv2cmdline(argc,argv);
uname(&uts);
getcwd(cwd,2000);
Progname = argv[0] ;
argc --;
argv++;
ErrorInit(NULL, NULL, NULL) ;
DiagInit(NULL, NULL, NULL) ;
if (argc == 0) usage_exit();
parse_commandline(argc, argv);
check_options();
if (checkoptsonly) return(0);
dump_options(stdout);
if(LogFile){
fpLog = fopen(LogFile,"w");
if(fpLog == NULL){
printf("ERROR: opening %s\n",LogFile);
exit(1);
}
dump_options(fpLog);
}
if(SynthSeed < 0) SynthSeed = PDFtodSeed();
srand48(SynthSeed);
SUBJECTS_DIR = getenv("SUBJECTS_DIR");
// Create output directory
printf("Creating %s\n",OutTop);
err = fio_mkdirp(OutTop,0777);
if(err) exit(1);
for(nthFWHM=0; nthFWHM < nFWHMList; nthFWHM++){
for(nthThresh = 0; nthThresh < nThreshList; nthThresh++){
for(nthSign = 0; nthSign < nSignList; nthSign++){
if(SignList[nthSign] == 0) signstr = "abs";
if(SignList[nthSign] == +1) signstr = "pos";
if(SignList[nthSign] == -1) signstr = "neg";
sprintf(tmpstr,"%s/fwhm%02d/%s/th%02d",
OutTop,(int)round(FWHMList[nthFWHM]),
signstr,(int)round(10*ThreshList[nthThresh]));
sprintf(fname,"%s/%s.csd",tmpstr,csdbase);
if(fio_FileExistsReadable(fname)){
printf("ERROR: output file %s exists\n",fname);
if(fpLog) fprintf(fpLog,"ERROR: output file %s exists\n",fname);
exit(1);
}
err = fio_mkdirp(tmpstr,0777);
if(err) exit(1);
}
}
}
// Load the target surface
sprintf(tmpstr,"%s/%s/surf/%s.%s",SUBJECTS_DIR,subject,hemi,surfname);
printf("Loading %s\n",tmpstr);
surf = MRISread(tmpstr);
if(!surf) return(1);
// Handle masking
if(LabelFile){
printf("Loading label file %s\n",LabelFile);
sprintf(tmpstr,"%s/%s/label/%s.%s.label",
SUBJECTS_DIR,subject,hemi,LabelFile);
if(!fio_FileExistsReadable(tmpstr)){
printf(" Cannot find label file %s\n",tmpstr);
sprintf(tmpstr,"%s",LabelFile);
printf(" Trying label file %s\n",tmpstr);
if(!fio_FileExistsReadable(tmpstr)){
printf(" ERROR: cannot read or find label file %s\n",LabelFile);
exit(1);
}
}
printf("Loading %s\n",tmpstr);
clabel = LabelRead(NULL, tmpstr);
mask = MRISlabel2Mask(surf, clabel, NULL);
FreeMask = 1;
}
if(MaskFile){
printf("Loading %s\n",MaskFile);
mask = MRIread(MaskFile);
if(mask == NULL) exit(1);
}
if(mask && SaveMask){
sprintf(tmpstr,"%s/mask.mgh",OutTop);
printf("Saving mask to %s\n",tmpstr);
err = MRIwrite(mask,tmpstr);
if(err) exit(1);
}
// Compute search space
searchspace = 0;
nmask = 0;
for(n=0; n < surf->nvertices; n++){
if(mask && MRIgetVoxVal(mask,n,0,0,0) < 0.5) continue;
searchspace += surf->vertices[n].area;
nmask++;
}
printf("Found %d voxels in mask\n",nmask);
if(surf->group_avg_surface_area > 0)
searchspace *= (surf->group_avg_surface_area/surf->total_area);
printf("search space %g mm2\n",searchspace);
avgvtxarea = searchspace/nmask;
printf("average vertex area %g mm2\n",avgvtxarea);
// Determine how many iterations are needed for each FWHM
nSmoothsList = (int *) calloc(sizeof(int),nFWHMList);
for(nthFWHM=0; nthFWHM < nFWHMList; nthFWHM++){
nSmoothsList[nthFWHM] = MRISfwhm2niters(FWHMList[nthFWHM], surf);
printf("%2d %5.1f %4d\n",nthFWHM,FWHMList[nthFWHM],nSmoothsList[nthFWHM]);
if(fpLog) fprintf(fpLog,"%2d %5.1f %4d\n",nthFWHM,FWHMList[nthFWHM],nSmoothsList[nthFWHM]);
}
printf("\n");
// Allocate the CSDs
for(nthFWHM=0; nthFWHM < nFWHMList; nthFWHM++){
for(nthThresh = 0; nthThresh < nThreshList; nthThresh++){
for(nthSign = 0; nthSign < nSignList; nthSign++){
csd = CSDalloc();
sprintf(csd->simtype,"%s","null-z");
sprintf(csd->anattype,"%s","surface");
sprintf(csd->subject,"%s",subject);
sprintf(csd->hemi,"%s",hemi);
sprintf(csd->contrast,"%s","NA");
csd->seed = SynthSeed;
csd->nreps = nRepetitions;
csd->thresh = ThreshList[nthThresh];
csd->threshsign = SignList[nthSign];
csd->nullfwhm = FWHMList[nthFWHM];
csd->varfwhm = -1;
csd->searchspace = searchspace;
CSDallocData(csd);
csdList[nthFWHM][nthThresh][nthSign] = csd;
}
}
}
// Alloc the z map
z = MRIallocSequence(surf->nvertices, 1,1, MRI_FLOAT, 1);
// Set up the random field specification
rfs = RFspecInit(SynthSeed,NULL);
rfs->name = strcpyalloc("gaussian");
rfs->params[0] = 0;
rfs->params[1] = 1;
printf("Thresholds (%d): ",nThreshList);
for(n=0; n < nThreshList; n++) printf("%5.2f ",ThreshList[n]);
printf("\n");
printf("Signs (%d): ",nSignList);
for(n=0; n < nSignList; n++) printf("%2d ",SignList[n]);
printf("\n");
printf("FWHM (%d): ",nFWHMList);
for(n=0; n < nFWHMList; n++) printf("%5.2f ",FWHMList[n]);
printf("\n");
// Start the simulation loop
printf("\n\nStarting Simulation over %d Repetitions\n",nRepetitions);
if(fpLog) fprintf(fpLog,"\n\nStarting Simulation over %d Repetitions\n",nRepetitions);
TimerStart(&mytimer) ;
for(nthRep = 0; nthRep < nRepetitions; nthRep++){
msecTime = TimerStop(&mytimer) ;
printf("%5d %7.1f ",nthRep,(msecTime/1000.0)/60);
if(fpLog) {
fprintf(fpLog,"%5d %7.1f ",nthRep,(msecTime/1000.0)/60);
fflush(fpLog);
}
// Synthesize an unsmoothed z map
RFsynth(z,rfs,mask);
nSmoothsPrev = 0;
// Loop through FWHMs
for(nthFWHM=0; nthFWHM < nFWHMList; nthFWHM++){
printf("%d ",nthFWHM);
if(fpLog) {
fprintf(fpLog,"%d ",nthFWHM);
fflush(fpLog);
}
nSmoothsDelta = nSmoothsList[nthFWHM] - nSmoothsPrev;
nSmoothsPrev = nSmoothsList[nthFWHM];
// Incrementally smooth z
MRISsmoothMRI(surf, z, nSmoothsDelta, mask, z); // smooth z
// Rescale
RFrescale(z,rfs,mask,z);
// Slightly tortured way to get the right p-values because
// RFstat2P() computes one-sided, but I handle sidedness
// during thresholding.
// First, use zabs to get a two-sided pval bet 0 and 0.5
zabs = MRIabs(z,zabs);
p = RFstat2P(zabs,rfs,mask,0,p);
// Next, mult pvals by 2 to get two-sided bet 0 and 1
MRIscalarMul(p,p,2.0);
sig = MRIlog10(p,NULL,sig,1); // sig = -log10(p)
for(nthThresh = 0; nthThresh < nThreshList; nthThresh++){
for(nthSign = 0; nthSign < nSignList; nthSign++){
csd = csdList[nthFWHM][nthThresh][nthSign];
// If test is not ABS then apply the sign
if(csd->threshsign != 0) MRIsetSign(sig,z,0);
// Get the max stats
sigmax = MRIframeMax(sig,0,mask,csd->threshsign,
&cmax,&rmax,&smax);
zmax = MRIgetVoxVal(z,cmax,rmax,smax,0);
if(csd->threshsign == 0){
zmax = fabs(zmax);
sigmax = fabs(sigmax);
}
// Mask
if(mask) MRImask(sig,mask,sig,0.0,0.0);
// Surface clustering
MRIScopyMRI(surf, sig, 0, "val");
if(csd->threshsign == 0) threshadj = csd->thresh;
else threshadj = csd->thresh - log10(2.0); // one-sided test
SurfClustList = sclustMapSurfClusters(surf,threshadj,-1,csd->threshsign,
0,&nClusters,NULL);
// Actual area of cluster with max area
csize = sclustMaxClusterArea(SurfClustList, nClusters);
// Number of vertices of cluster with max number of vertices.
// Note: this may be a different cluster from above!
csizen = sclustMaxClusterCount(SurfClustList, nClusters);
// Area of this cluster based on average vertex area. This just scales
// the number of vertices.
csizeavg = csizen * avgvtxarea;
if(UseAvgVtxArea) csize = csizeavg;
// Store results
csd->nClusters[nthRep] = nClusters;
csd->MaxClusterSize[nthRep] = csize;
csd->MaxSig[nthRep] = sigmax;
csd->MaxStat[nthRep] = zmax;
} // Sign
} // Thresh
} // FWHM
printf("\n");
if(fpLog) fprintf(fpLog,"\n");
if(SaveEachIter || fio_FileExistsReadable(SaveFile)) SaveOutput();
if(fio_FileExistsReadable(StopFile)) {
printf("Found stop file %s\n",StopFile);
goto finish;
}
} // Simulation Repetition
finish:
SaveOutput();
msecTime = TimerStop(&mytimer) ;
printf("Total Sim Time %g min (%g per rep)\n",
msecTime/(1000*60.0),(msecTime/(1000*60.0))/nthRep);
if(fpLog) fprintf(fpLog,"Total Sim Time %g min (%g per rep)\n",
msecTime/(1000*60.0),(msecTime/(1000*60.0))/nthRep);
if(DoneFile){
fp = fopen(DoneFile,"w");
fprintf(fp,"%g\n",msecTime/(1000*60.0));
fclose(fp);
}
printf("mri_mcsim done\n");
if(fpLog){
fprintf(fpLog,"mri_mcsim done\n");
fclose(fpLog);
}
exit(0);
}
void
avtZoneDumpFilter::PostExecute()
{
// skip dump if not enabled
if(!atts.GetEnabled())
return;
#ifdef PARALLEL
// loop index
int i;
// get the number of processors and the current processor id
int nprocs = PAR_Size();
int procid = PAR_Rank();
// get the number of zone infos to send
int n_snd_zones = zones.size();
// size of each zone info
int zinfo_size = ZoneInfo::PackedSize();
// calculate the total send message size
int snd_msg_size = n_snd_zones * zinfo_size;
// send buffer
unsigned char *snd_msg = NULL;
// vars for the root processor
// size of the gather message
int rcv_msg_size = 0;
// holds size of the msg from each other processor
int *rcv_count = NULL;
// holds the displacement of the msg from each other processor
int *rcv_disp = NULL;
// receive buffer
unsigned char *rcv_msg = NULL;
if(procid == 0)
{
// allocate space for these for root proc only
rcv_count = new int[nprocs];
rcv_disp = new int[nprocs];
}
// gather message sizes from all procs to root proc
MPI_Gather(&snd_msg_size,1, MPI_INT,
rcv_count, 1, MPI_INT,
0, VISIT_MPI_COMM);
// find message offsets and total rcv size
if(procid == 0)
{
rcv_disp[0] = 0;
rcv_msg_size = rcv_count[0];
for( i=1; i<nprocs;i++)
{
rcv_disp[i] = rcv_count[i-1] + rcv_disp[i-1];
rcv_msg_size += rcv_count[i];
}
}
// get total # of zone infos
int nrcv_zones = rcv_msg_size / zinfo_size;
// create msg to send to the root proc
if(snd_msg_size > 0)
{
snd_msg= new unsigned char[snd_msg_size];
unsigned char *snd_msg_ptr = snd_msg;
// pack zone infos
for(i=0; i < n_snd_zones; i++)
{
zones[i].Pack(snd_msg_ptr);
snd_msg_ptr+= zinfo_size;
}
}
if(procid == 0 && rcv_msg_size > 0)
{
// create the rcv buffer for the root proc
rcv_msg = new unsigned char[rcv_msg_size];
}
// gather all zone infos
MPI_Gatherv(snd_msg, snd_msg_size, MPI_UNSIGNED_CHAR,
rcv_msg, rcv_count, rcv_disp, MPI_UNSIGNED_CHAR,
0,VISIT_MPI_COMM);
if(procid == 0 )
{
// unpack all rcvd zones
std::vector<ZoneInfo> rcv_zones(nrcv_zones);
unsigned char *rcv_msg_ptr = rcv_msg;
for( i = 0; i < nrcv_zones; i++)
{
rcv_zones[i].Unpack(rcv_msg_ptr);
rcv_msg_ptr += zinfo_size;
}
// save all zones
SaveOutput(atts.GetOutputFile(),rcv_zones);
}
// cleanup
if(snd_msg)
delete[] snd_msg;
if(rcv_msg)
delete[] rcv_msg;
if(rcv_count)
delete[] rcv_count;
if(rcv_disp)
delete[] rcv_disp;
#else
// for serial case, simply dump out zones found during the exe pass.
SaveOutput(atts.GetOutputFile(),zones);
#endif
}
void mexFunction( int nlhs, mxArray *plhs[],
int nrhs, const mxArray *prhs[] )
{
float Score;
float BestScore = HugeScore;
int p, i;
SetupParams(nrhs,prhs);
clock_t Clock0;
KK K1; // main KK class, for all data
K1.penaltyMix = PenaltyMix;
Clock0 = clock(); // start timer
K1.LoadData(prhs[0]); // load .fet file
mwSize dim1[2] = {1,K1.nPoints};
plhs[0] = mxCreateNumericArray(2, dim1, mxDOUBLE_CLASS, mxREAL);
double *OutputArray = mxGetPr(plhs[0]);
// Seed random number generator
srand(RandomSeed);
// open distance dump file if required
//if (DistDump) Distfp = fopen("DISTDUMP", "w");
// start with provided file, if required
if (!mxIsEmpty(prhs[1])) {
Output("Starting from existing clusters \n");
BestScore = K1.CEM(prhs[1], 1, 1);
Output("%d->%d Clusters: Score %f\n\n", K1.nStartingClusters, K1.nClustersAlive, BestScore);
for(p=0; p<K1.nPoints; p++) K1.BestClass[p] = K1.Class[p];
SaveOutput(K1.BestClass, OutputArray);
}
// loop through numbers of clusters ...
for(K1.nStartingClusters=MinClusters; K1.nStartingClusters<=MaxClusters; K1.nStartingClusters++) for(i=0; i<nStarts; i++) {
// do CEM iteration
Output("Starting from %d clusters...\n", K1.nStartingClusters);
Score = K1.Cluster();
Output("%d->%d Clusters: Score %f, best is %f\n", K1.nStartingClusters, K1.nClustersAlive, Score, BestScore);
if (Score < BestScore) {
Output("THE BEST YET!\n");
// New best classification found
BestScore = Score;
for(p=0; p<K1.nPoints; p++) K1.BestClass[p] = K1.Class[p];
SaveOutput(K1.BestClass, OutputArray);
}
Output("\n");
}
SaveOutput(K1.BestClass, OutputArray);
Output("That took %f seconds.\n", (clock()-Clock0)/(float) CLOCKS_PER_SEC);
//if (DistDump) fclose(Distfp);
}
void merge(TString outfile, TString firstfile, TString pat=defaultPattern)
{
TList decayModes;
bool save=false;
TString dir=".";
TString first="";
long flags=0,id=0,size=0,modtime=0;
if(firstfile.Contains("~"))
{
cout<<"ERROR: input path cannot contain '~' character."<<endl;
return;
}
// Check pattern
TRegexp pattern(pat,true);
if(pattern.Status()!=0)
{
cout<<"ERROR: Wrong regular expression."<<endl;
return;
}
// Load libraries
loadLibs();
gSystem->GetPathInfo(firstfile, &id, &size, &flags, &modtime);
bool isDirectory = flags&2;
if(isDirectory)
{
dir=firstfile;
first="";
}
else
{
//Separate directory from filename
int separator=firstfile.Last('/');
if(separator!=-1) dir=firstfile(0,separator+1);
first=firstfile(separator+1,firstfile.Length());
}
cout<<"Output file: "<<outfile<<endl;
cout<<"Input dir: "<<dir<<endl;
if(!isDirectory) cout<<"First file: "<<first<<endl;
cout<<endl;
// Get file list and add first file if present
TList *files = getFileList(first,dir,pattern);
if(!isDirectory)
{
TFile *ff = new TFile(firstfile,"READ");
files->AddBefore(files->First(),ff);
}
// Merge files
TIter next(files);
TFile *file;
while(file = (TFile*)next())
{
if(!file->IsOpen()) continue;
ReadFile(file,decayModes);
cout<<"=============================="<<endl;
save=true;
}
// Save output
if(save)
{
cout<<"Saving..."<<endl;
SaveOutput(outfile,decayModes);
cout<<"Output saved to "<<outfile<<endl<<endl;
}
else cout<<"Nothing to save."<<endl<<endl;
// Closing files
cout<<"Closing files..."<<endl;
TIter cl(files);
while(file = (TFile*)cl()) { file->Close(); delete file; }
delete files;
gSystem->Exit(0);
}
// CEM(StartFile) - Does a whole CEM algorithm from a random start
// optional start file loads this cluster file to start iteration
// if Recurse is 0, it will not try and split.
// if InitRand is 0, use cluster assignments already in structure
float KK::CEM(char *CluFile /*= NULL*/, int Recurse /*=1*/, int InitRand /*=1*/) {
int p, c, i;
int nChanged;
int Iter;
Array<int> OldClass(nPoints);
float Score, OldScore;
int LastStepFull; // stores whether the last step was a full one
int DidSplit;
if (CluFile && *CluFile) LoadClu(CluFile);
else if (InitRand) {
// initialize data to random
if (nStartingClusters>1)
for(p=0; p<nPoints; p++) Class[p] = irand(1, nStartingClusters-1);
else
for(p=0; p<nPoints; p++) Class[p] = 0;
for(c=0; c<MaxPossibleClusters; c++) ClassAlive[c] = (c<nStartingClusters);
}
// set all clases to alive
Reindex();
// main loop
Iter = 0;
FullStep = 1;
do {
// Store old classifications
for(p=0; p<nPoints; p++) OldClass[p] = Class[p];
// M-step - calculate class weights, means, and covariance matrices for each class
MStep();
// E-step - calculate scores for each point to belong to each class
EStep();
// dump distances if required
if (DistDump) MatPrint(Distfp, LogP.m_Data, DistDump, MaxPossibleClusters);
// C-step - choose best class for each
CStep();
// Would deleting any classes improve things?
if(Recurse) ConsiderDeletion();
// Calculate number changed
nChanged = 0;
for(p=0; p<nPoints; p++) nChanged += (OldClass[p] != Class[p]);
// Calculate score
OldScore = Score;
Score = ComputeScore();
if(Verbose>=1) {
if(Recurse==0) Output("\t");
Output("Iteration %d%c: %d clusters Score %.7g nChanged %d\n",
Iter, FullStep ? 'F' : 'Q', nClustersAlive, Score, nChanged);
}
Iter++;
if (Debug) {
for(p=0;p<nPoints;p++) BestClass[p] = Class[p];
SaveOutput(BestClass);
Output("Press return");
getchar();
}
// Next step a full step?
LastStepFull = FullStep;
FullStep = (
nChanged>ChangedThresh*nPoints
|| nChanged == 0
|| Iter%FullStepEvery==0
// || Score > OldScore Doesn't help!
// Score decreases are not because of quick steps!
) ;
if (Iter>MaxIter) {
Output("Maximum iterations exceeded\n");
break;
}
// try splitting
if ((Recurse && SplitEvery>0) && (Iter%SplitEvery==SplitEvery-1 || (nChanged==0 && LastStepFull))) {
DidSplit = TrySplits();
} else DidSplit = 0;
} while (nChanged > 0 || !LastStepFull || DidSplit);
if (DistDump) fprintf(Distfp, "\n");
return Score;
}
