Add options for text line height and normalization style

clstmhl.h

@@ -146,8 +146,9 @@ struct CLSTMText {
 struct CLSTMOCR {
   shared_ptr<INormalizer> normalizer;
   Network net;
-  int target_height = 48;
+  int target_height;// = 48; // to avoid unwanted values.
   int nclasses = -1;
+  string dewarp;  // Option for text-line normalization
   Sequence aligned, targets;
   Tensor2 image;
   void setLearningRate(float lr, float mom) { net->setLearningRate(lr, mom); }
@@ -161,7 +162,7 @@ struct CLSTMOCR {
       return false;
     }
     nclasses = net->codec.size();
-    normalizer.reset(make_CenterNormalizer());
+    normalizer.reset(make_Normalizer(dewarp));
     normalizer->target_height = target_height;
     return true;
   }
@@ -194,7 +195,7 @@ struct CLSTMOCR {
                             {"nhidden", nhidden}});
     net->initialize();
     net->codec.set(codec);
-    normalizer.reset(make_CenterNormalizer());
+    normalizer.reset(make_Normalizer(dewarp));
     normalizer->target_height = target_height;
   }
   std::wstring fwdbwd(TensorMap2 raw, const std::wstring &target) {

clstmocr.cc

@@ -46,8 +46,10 @@ int main1(int argc, char **argv) {
   string load_name = getsenv("load", "");
   if (load_name == "") THROW("must give load= parameter");
   CLSTMOCR clstm;
+  clstm.target_height = int(getrenv("target_height", 45));
+  clstm.dewarp = getsenv("dewarp", "none");
   clstm.load(load_name);
-
+  
   bool conf = getienv("conf", 0);
   string output = getsenv("output", "text");
   bool save_text = getienv("save_text", 1);

clstmocrtrain.cc

@@ -65,6 +65,19 @@ struct Dataset {
     for (auto s : fnames) gtnames.push_back(basename(s) + ".gt.txt");
     codec.build(gtnames, charsep);
   }
+  void getCodec(Codec &codec, vector<string> file_lists) {
+    // get codec from several files, including training files, validation files, 
+    // and perhaps testing files, in order to avoid unrecognized codecs
+    vector<string> gts;
+    for (int i=0; i<file_lists.size(); i++) {
+      vector<string> temp_names;
+      read_lines(temp_names, file_lists[i]);
+      for (auto s : temp_names) gts.push_back(basename(s) + ".gt.txt");
+    }
+    // build the codecs
+    codec.build(gts, charsep);
+  }
+
   void readSample(Tensor2 &raw, wstring &gt, int index) {
     string fname = fnames[index];
     string base = basename(fname);
@@ -92,12 +105,19 @@ int main1(int argc, char **argv) {
   int ntrain = getienv("ntrain", 10000000);
   string save_name = getsenv("save_name", "_ocr");
   int report_time = getienv("report_time", 0);
+  // vector storing the training and testing files
+  vector<string> file_lists;
 
   if (argc < 2 || argc > 3) THROW("... training [testing]");
   Dataset trainingset(argv[1]);
+  file_lists.push_back(argv[1]);
   assert(trainingset.size() > 0);
   Dataset testset;
-  if (argc > 2) testset.readFileList(argv[2]);
+  if (argc > 2) {
+    testset.readFileList(argv[2]);
+    file_lists.push_back(argv[2]);
+  }
+
   print("got", trainingset.size(), "files,", testset.size(), "tests");
 
   string load_name = getsenv("load", "");
@@ -108,13 +128,16 @@ int main1(int argc, char **argv) {
     clstm.load(load_name);
   } else {
     Codec codec;
-    trainingset.getCodec(codec);
+    //trainingset.getCodec(codec);
+    trainingset.getCodec(codec, file_lists);    // use all ground truth files
     print("got", codec.size(), "classes");
 
-    clstm.target_height = int(getrenv("target_height", 48));
+    clstm.target_height = int(getrenv("target_height", 45));
+    clstm.dewarp = getsenv("dewarp", "none");
     clstm.createBidi(codec.codec, getienv("nhidden", 100));
     clstm.setLearningRate(getdenv("lrate", 1e-4), getdenv("momentum", 0.9));
   }
+  file_lists.clear();   // clear the file_lists vector
   network_info(clstm.net);
 
   double test_error = 9999.0;
@@ -135,12 +158,16 @@ int main1(int argc, char **argv) {
   Trigger report_trigger(getienv("report_every", 100), ntrain, start);
   Trigger display_trigger(getienv("display_every", 0), ntrain, start);
 
+  double train_errors = 0.0;
+  double train_count = 0.0;
   for (int trial = start; trial < ntrain; trial++) {
     int sample = lrand48() % trainingset.size();
     Tensor2 raw;
     wstring gt;
     trainingset.readSample(raw, gt, sample);
     wstring pred = clstm.train(raw(), gt);
+    train_count += gt.size();
+    train_errors += levenshtein(pred, gt);
 
     if (report_trigger(trial)) {
       print(trial);
@@ -168,6 +195,15 @@ int main1(int argc, char **argv) {
       double count = tse.second;
       test_error = errors / count;
       print("ERROR", trial, test_error, "   ", errors, count);
+      double train_error;
+      if (train_errors > 0)
+         train_error = train_count / train_errors;
+      else
+         train_error = 9999.0;
+      print("Train ERROR: ", train_error);
+      train_count = 0.0;
+      train_errors = 0.0;
+
       if (test_error < best_error) {
         best_error = test_error;
         string fname = save_name + ".clstm";

extras.h

@@ -31,7 +31,7 @@ using std::min;
 // text line normalization
 
 struct INormalizer {
-  int target_height = 48;
+  int target_height; // = 48;
   float smooth2d = 1.0;
   float smooth1d = 0.3;
   float range = 4.0;