Add new option -l for combine_tessdata to list LSTM information

doc/combine_tessdata.1.asc

@@ -66,6 +66,9 @@ OPTIONS
 *-e* '.traineddata' 'FILE'...:
     Extracts the specified components from the .traineddata file
 
+*-l* '.traineddata' 'FILE'...:
+   List the network information.
+
 *-o* '.traineddata' 'FILE'...:
     Overwrites the specified components of the .traineddata file
     with those provided on the command line.

src/ccmain/tessedit.cpp

@@ -175,7 +175,7 @@ bool Tesseract::init_tesseract_lang_data(
       tessedit_ocr_engine_mode == OEM_TESSERACT_LSTM_COMBINED) {
 #endif  // ndef DISABLED_LEGACY_ENGINE
     if (mgr->IsComponentAvailable(TESSDATA_LSTM)) {
-      lstm_recognizer_ = new LSTMRecognizer(language_data_path_prefix);
+      lstm_recognizer_ = new LSTMRecognizer(language_data_path_prefix.c_str());
       ASSERT_HOST(lstm_recognizer_->Load(
           this->params(), lstm_use_matrix ? language : nullptr, mgr));
     } else {

src/ccutil/serialis.cpp

@@ -109,6 +109,24 @@ template TESS_API bool TFile::DeSerialize(std::vector<int32_t>& data);
 template TESS_API bool TFile::Serialize(const std::vector<double>& data);
 template TESS_API bool TFile::Serialize(const std::vector<int32_t>& data);
 
+bool TFile::DeSerialize(std::string& data) {
+  uint32_t size;
+  if (!DeSerialize(&size)) {
+    return false;
+  } else if (size > 0) {
+    // TODO: optimize.
+    data.resize(size);
+    return DeSerialize(&data[0], size);
+  }
+  data.clear();
+  return true;
+}
+
+bool TFile::Serialize(const std::string& data) {
+  uint32_t size = data.size();
+  return Serialize(&size) && Serialize(data.c_str(), size);
+}
+
 bool TFile::DeSerialize(std::vector<char>& data) {
   uint32_t size;
   if (!DeSerialize(&size)) {

src/ccutil/serialis.h

@@ -85,6 +85,7 @@ class TESS_API TFile {
   }
 
   // Deserialize data.
+  bool DeSerialize(std::string& data);
   bool DeSerialize(std::vector<char>& data);
   template <typename T> bool DeSerialize(std::vector<T>& data);
   template <typename T>
@@ -93,6 +94,7 @@ class TESS_API TFile {
   }
 
   // Serialize data.
+  bool Serialize(const std::string& data);
   bool Serialize(const std::vector<char>& data);
   template <typename T> bool Serialize(const std::vector<T>& data);
   template <typename T>

src/lstm/lstmrecognizer.cpp

@@ -47,7 +47,7 @@ const double kDictRatio = 2.25;
 // Default certainty offset to give the dictionary a chance.
 const double kCertOffset = -0.085;
 
-LSTMRecognizer::LSTMRecognizer(const STRING language_data_path_prefix)
+LSTMRecognizer::LSTMRecognizer(const char* language_data_path_prefix)
     : LSTMRecognizer::LSTMRecognizer() {
   ccutil_.language_data_path_prefix = language_data_path_prefix;
 }
@@ -90,7 +90,7 @@ bool LSTMRecognizer::Serialize(const TessdataManager* mgr, TFile* fp) const {
                           !mgr->IsComponentAvailable(TESSDATA_LSTM_UNICHARSET);
   if (!network_->Serialize(fp)) return false;
   if (include_charsets && !GetUnicharset().save_to_file(fp)) return false;
-  if (!network_str_.Serialize(fp)) return false;
+  if (!fp->Serialize(network_str_)) return false;
   if (!fp->Serialize(&training_flags_)) return false;
   if (!fp->Serialize(&training_iteration_)) return false;
   if (!fp->Serialize(&sample_iteration_)) return false;
@@ -112,7 +112,7 @@ bool LSTMRecognizer::DeSerialize(const TessdataManager* mgr, TFile* fp) {
                           !mgr->IsComponentAvailable(TESSDATA_LSTM_UNICHARSET);
   if (include_charsets && !ccutil_.unicharset.load_from_file(fp, false))
     return false;
-  if (!network_str_.DeSerialize(fp)) return false;
+  if (!fp->DeSerialize(network_str_)) return false;
   if (!fp->DeSerialize(&training_flags_)) return false;
   if (!fp->DeSerialize(&training_iteration_)) return false;
   if (!fp->DeSerialize(&sample_iteration_)) return false;

src/lstm/lstmrecognizer.h

@@ -53,13 +53,20 @@ enum TrainingFlags {
 class TESS_API LSTMRecognizer {
  public:
   LSTMRecognizer();
-  LSTMRecognizer(const STRING language_data_path_prefix);
+  LSTMRecognizer(const char* language_data_path_prefix);
   ~LSTMRecognizer();
 
   int NumOutputs() const { return network_->NumOutputs(); }
+
+  // Return the training iterations.
   int training_iteration() const { return training_iteration_; }
+
+  // Return the sample iterations.
   int sample_iteration() const { return sample_iteration_; }
-  double learning_rate() const { return learning_rate_; }
+
+  // Return the learning rate.
+  float learning_rate() const { return learning_rate_; }
+
   LossType OutputLossType() const {
     if (network_ == nullptr) return LT_NONE;
     StaticShape shape;
@@ -101,6 +108,22 @@ class TESS_API LSTMRecognizer {
       return learning_rate_;
     }
   }
+
+  // Return the network string.
+  const char* GetNetwork() const {
+    return network_str_.c_str();
+  }
+
+  // Return the adam beta.
+  float GetAdamBeta() const {
+    return adam_beta_;
+  }
+
+  // Return the momentum.
+  float GetMomentum() const {
+    return momentum_;
+  }
+
   // Multiplies the all the learning rate(s) by the given factor.
   void ScaleLearningRate(double factor) {
     ASSERT_HOST(network_ != nullptr && network_->type() == NT_SERIES);
@@ -142,6 +165,8 @@ class TESS_API LSTMRecognizer {
   void SetIteration(int iteration) { sample_iteration_ = iteration; }
   // Accessors for textline image normalization.
   int NumInputs() const { return network_->NumInputs(); }
+
+  // Return the null char index.
   int null_char() const { return null_char_; }
 
   // Loads a model from mgr, including the dictionary only if lang is not null.
@@ -268,7 +293,7 @@ class TESS_API LSTMRecognizer {
   UnicharCompress recoder_;
 
   // ==Training parameters that are serialized to provide a record of them.==
-  STRING network_str_;
+  std::string network_str_;
   // Flags used to determine the training method of the network.
   // See enum TrainingFlags above.
   int32_t training_flags_;

src/training/combine_tessdata.cpp

@@ -22,6 +22,7 @@
 #include "tessdatamanager.h"
 
 #include <cerrno>
+#include <iostream>             // std::cout
 
 using namespace tesseract;
 
@@ -174,6 +175,30 @@ int main(int argc, char **argv) {
   } else if (argc == 3 && strcmp(argv[1], "-d") == 0) {
     // Initialize TessdataManager with the data in the given traineddata file.
     tm.Init(argv[2]);
+  } else if (argc == 3 && strcmp(argv[1], "-l") == 0) {
+    if (!tm.Init(argv[2])) {
+      tprintf("Failed to read %s\n", argv[2]);
+      return EXIT_FAILURE;
+    }
+    tesseract::TFile fp;
+    if (tm.GetComponent(tesseract::TESSDATA_LSTM, &fp)) {
+      tesseract::LSTMRecognizer recognizer;
+      if (!recognizer.DeSerialize(&tm, &fp)) {
+        tprintf("Failed to deserialize LSTM in %s!\n", argv[2]);
+        return EXIT_FAILURE;
+      }
+      std::cout << "LSTM: network=" << recognizer.GetNetwork()
+                << ", int_mode=" << recognizer.IsIntMode()
+                << ", recoding=" << recognizer.IsRecoding()
+                << ", iteration=" << recognizer.training_iteration()
+                << ", sample_iteration=" << recognizer.sample_iteration()
+                << ", null_char=" << recognizer.null_char()
+                << ", learning_rate=" << recognizer.learning_rate()
+                << ", momentum=" << recognizer.GetMomentum()
+                << ", adam_beta=" << recognizer.GetAdamBeta()
+                << '\n';
+    }
+    return EXIT_SUCCESS;
   } else {
     printf("Usage for combining tessdata components:\n"
            "  %s language_data_path_prefix\n"
@@ -188,10 +213,13 @@ int main(int argc, char **argv) {
            argv[0], argv[0]);
     printf("Usage for unpacking all tessdata components:\n"
            "  %s -u traineddata_file output_path_prefix\n"
-           "  (e.g. %s -u eng.traineddata tmp/eng.)\n", argv[0], argv[0]);
+           "  (e.g. %s -u eng.traineddata tmp/eng.)\n\n", argv[0], argv[0]);
+    printf("Usage for listing the network information\n"
+           "  %s -l traineddata_file\n"
+           "  (e.g. %s -l eng.traineddata)\n\n", argv[0], argv[0]);
     printf(
         "Usage for listing directory of components:\n"
-        "  %s -d traineddata_file\n",
+        "  %s -d traineddata_file\n\n",
         argv[0]);
     printf(
         "Usage for compacting LSTM component to int:\n"