changeset 39:39499ccf7656 tony

Return candidates as a stack of values per frame

--- a/LocalCandidatePYIN.cpp	Mon Jan 27 15:26:32 2014 +0000
+++ b/LocalCandidatePYIN.cpp	Tue Jan 28 11:51:47 2014 +0000
@@ -25,9 +25,11 @@
 // #include <iostream>
 #include <cmath>
 #include <complex>
+#include <map>
 
 using std::string;
 using std::vector;
+using std::map;
 using Vamp::RealTime;
 
 
@@ -222,16 +224,14 @@
     d.description = "Multiple candidate pitch tracks.";
     d.unit = "Hz";
     d.hasFixedBinCount = false;
-    // d.binCount = 1;
     d.hasKnownExtents = true;
     d.minValue = m_fmin;
-    d.maxValue = 500;
+    d.maxValue = 500; //!!!???
     d.isQuantized = false;
     d.sampleType = OutputDescriptor::FixedSampleRate;
     d.sampleRate = (m_inputSampleRate / m_stepSize);
     d.hasDuration = false;
     outputs.push_back(d);
-    // m_oPitchTrackCandidates = outputNumber++;
 
     return outputs;
 }
@@ -279,7 +279,6 @@
 LocalCandidatePYIN::process(const float *const *inputBuffers, RealTime timestamp)
 {
     timestamp = timestamp + Vamp::RealTime::frame2RealTime(m_blockSize/4, lrintf(m_inputSampleRate));
-    FeatureSet fs;
     
     double *dInputBuffers = new double[m_blockSize];
     for (size_t i = 0; i < m_blockSize; ++i) dInputBuffers[i] = inputBuffers[0][i];
@@ -318,22 +317,19 @@
     }
     m_timestamp.push_back(timestamp);
 
-    return fs;
+    return FeatureSet();
 }
 
 LocalCandidatePYIN::FeatureSet
 LocalCandidatePYIN::getRemainingFeatures()
 {
-    FeatureSet fs;
-    Feature f;
-    f.hasTimestamp = true;
-    f.hasDuration = false;
-    f.values.push_back(0);
+    // timestamp -> candidate number -> value
+    map<RealTime, map<int, float> > featureValues;
 
     std::cerr << "in remaining features" << std::endl;
 
     if (m_pitchProb.empty()) {
-        return fs;
+        return FeatureSet();
     }
 
     // MONO-PITCH STUFF
@@ -378,6 +374,9 @@
         }
     }
 
+    map<int, int> candidateActuals;
+    map<int, std::string> candidateLabels;
+
     int actualCandidateNumber = 0;
     for (size_t iCandidate = 0; iCandidate < m_nCandidate; ++iCandidate) {
         bool isDuplicate = false;
@@ -392,15 +391,15 @@
         {
             std::ostringstream convert;
             convert << actualCandidateNumber++;
-            f.label = convert.str();
+            candidateLabels[iCandidate] = convert.str();
+            candidateActuals[iCandidate] = actualCandidateNumber;
             std::cerr << freqNumber[iCandidate] << " " << freqMean[iCandidate] << std::endl;
             for (size_t iFrame = 0; iFrame < nFrame; ++iFrame) 
             {
                 if (pitchTracks[iCandidate][iFrame] > 0)
                 {
-                    f.values[0] = pitchTracks[iCandidate][iFrame];
-                    f.timestamp = m_timestamp[iFrame];
-                    fs[m_oPitchTrackCandidates].push_back(f);
+                    featureValues[m_timestamp[iFrame]][iCandidate] = 
+                        pitchTracks[iCandidate][iFrame];
                 }
             }
         }
@@ -408,6 +407,31 @@
     }
 
     // only retain those that are close to their means
+    //!!!??? what does the above mean?
+
+    // adapt our features so as to return a stack of candidate values
+    // per frame
+
+    FeatureSet fs;
+
+    for (map<RealTime, map<int, float> >::const_iterator i =
+             featureValues.begin(); i != featureValues.end(); ++i) {
+        Feature f;
+        f.hasTimestamp = true;
+        f.timestamp = i->first;
+        int nextCandidate = candidateActuals.begin()->second;
+        for (map<int, float>::const_iterator j = 
+                 i->second.begin(); j != i->second.end(); ++j) {
+            while (candidateActuals[j->first] > nextCandidate) {
+                f.values.push_back(0);
+                ++nextCandidate;
+            }
+            f.values.push_back(j->second);
+            nextCandidate = j->first + 1;
+        }
+        //!!! can't use labels?
+        fs[0].push_back(f);
+    }
 
     return fs;
 }