automatic commit

[folded-ctf.git] / parsing.cc

diff --git a/parsing.cc b/parsing.cc
new file mode 100644 (file)
index 0000000..9d0060d
--- /dev/null
+++ b/parsing.cc
@@ -0,0 +1,186 @@
+
+///////////////////////////////////////////////////////////////////////////
+// This program is free software: you can redistribute it and/or modify  //
+// it under the terms of the version 3 of the GNU General Public License //
+// as published by the Free Software Foundation.                         //
+//                                                                       //
+// This program is distributed in the hope that it will be useful, but   //
+// WITHOUT ANY WARRANTY; without even the implied warranty of            //
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU      //
+// General Public License for more details.                              //
+//                                                                       //
+// You should have received a copy of the GNU General Public License     //
+// along with this program. If not, see <http://www.gnu.org/licenses/>.  //
+//                                                                       //
+// Written by Francois Fleuret, (C) IDIAP                                //
+// Contact <francois.fleuret@idiap.ch> for comments & bug reports        //
+///////////////////////////////////////////////////////////////////////////
+
+#include "parsing.h"
+#include "fusion_sort.h"
+
+Parsing::Parsing(LabelledImagePool *image_pool,
+                 PoseCellHierarchy *hierarchy,
+                 scalar_t proportion_negative_cells,
+                 int image_index) {
+
+  _image_pool = image_pool;
+  _image_index = image_index;
+
+  PoseCellSet cell_set;
+  LabelledImage *image;
+
+  image = _image_pool->grab_image(_image_index);
+
+  hierarchy->add_root_cells(image, &cell_set);
+
+  int *kept = new int[cell_set.nb_cells()];
+
+  _nb_cells = 0;
+
+  for(int c = 0; c < cell_set.nb_cells(); c++) {
+    int l = image->pose_cell_label(cell_set.get_cell(c));
+    kept[c] = (l > 0) || (l < 0 && drand48() < proportion_negative_cells);
+    if(kept[c]) _nb_cells++;
+  }
+
+  _cells = new PoseCell[_nb_cells];
+  _responses = new scalar_t[_nb_cells];
+  _labels = new int[_nb_cells];
+  _nb_positives = 0;
+  _nb_negatives = 0;
+
+  int d = 0;
+  for(int c = 0; c < cell_set.nb_cells(); c++) {
+    if(kept[c]) {
+      _cells[d] = *(cell_set.get_cell(c));
+      _labels[d] = image->pose_cell_label(&_cells[d]);
+      _responses[d] = 0;
+      if(_labels[d] < 0) {
+        _nb_negatives++;
+      } else if(_labels[d] > 0) {
+        _nb_positives++;
+      }
+      d++;
+    }
+  }
+
+  delete[] kept;
+
+  _image_pool->release_image(_image_index);
+}
+
+Parsing::~Parsing() {
+  delete[] _cells;
+  delete[] _responses;
+  delete[] _labels;
+}
+
+void Parsing::down_one_level(PoseCellHierarchy *hierarchy,
+                             int level, int *sample_nb_occurences, scalar_t *sample_responses) {
+  PoseCellSet cell_set;
+  LabelledImage *image;
+
+  int new_nb_cells = 0;
+  for(int c = 0; c < _nb_cells; c++) {
+    new_nb_cells += sample_nb_occurences[c];
+  }
+
+  PoseCell *new_cells = new PoseCell[new_nb_cells];
+  scalar_t *new_responses = new scalar_t[new_nb_cells];
+  int *new_labels = new int[new_nb_cells];
+
+  image = _image_pool->grab_image(_image_index);
+  int b = 0;
+
+  for(int c = 0; c < _nb_cells; c++) {
+
+    if(sample_nb_occurences[c] > 0) {
+
+      cell_set.erase_content();
+      hierarchy->add_subcells(level, _cells + c, &cell_set);
+
+      if(_labels[c] > 0) {
+        ASSERT(sample_nb_occurences[c] == 1);
+        int e = -1;
+        for(int d = 0; d < cell_set.nb_cells(); d++) {
+          if(image->pose_cell_label(cell_set.get_cell(d)) > 0) {
+            ASSERT(e < 0);
+            e = d;
+          }
+        }
+        ASSERT(e >= 0);
+        ASSERT(b < new_nb_cells);
+        new_cells[b] = *(cell_set.get_cell(e));
+        new_responses[b] = sample_responses[c];
+        new_labels[b] = 1;
+        b++;
+      }
+
+      else if(_labels[c] < 0) {
+        for(int d = 0; d < sample_nb_occurences[c]; d++) {
+          ASSERT(b < new_nb_cells);
+          new_cells[b] = *(cell_set.get_cell(int(drand48() * cell_set.nb_cells())));
+          new_responses[b] = sample_responses[c];
+          new_labels[b] = -1;
+          b++;
+        }
+      }
+
+      else {
+        cerr << "INCONSISTENCY" << endl;
+        abort();
+      }
+    }
+  }
+
+  ASSERT(b == new_nb_cells);
+
+  _image_pool->release_image(_image_index);
+
+  delete[] _cells;
+  delete[] _labels;
+  delete[] _responses;
+  _nb_cells = new_nb_cells;
+  _cells = new_cells;
+  _labels = new_labels;
+  _responses = new_responses;
+}
+
+void Parsing::update_cell_responses(PiFeatureFamily *pi_feature_family,
+                                    Classifier *classifier) {
+  LabelledImage *image;
+
+  image = _image_pool->grab_image(_image_index);
+  image->compute_rich_structure();
+
+  SampleSet *samples = new SampleSet(pi_feature_family->nb_features(), 1);
+
+  for(int c = 0; c < _nb_cells; c++) {
+    samples->set_sample(0, pi_feature_family, image, &_cells[c], 0);
+    _responses[c] += classifier->response(samples, 0);
+    ASSERT(!isnan(_responses[c]));
+  }
+
+  _image_pool->release_image(_image_index);
+  delete samples;
+}
+
+void Parsing::collect_samples(SampleSet *samples,
+                              PiFeatureFamily *pi_feature_family,
+                              int s,
+                              int *to_collect) {
+  LabelledImage *image;
+
+  image = _image_pool->grab_image(_image_index);
+  image->compute_rich_structure();
+
+  for(int c = 0; c < _nb_cells; c++) {
+    if(to_collect[c]) {
+      samples->set_sample(s, pi_feature_family, image, &_cells[c], _labels[c]);
+      s++;
+    }
+  }
+
+  _image_pool->release_image(_image_index);
+}