tesseract  4.1.1
tesseract::LSTMTester Class Reference

#include <lstmtester.h>

Public Member Functions

 LSTMTester (int64_t max_memory)
 
bool LoadAllEvalData (const STRING &filenames_file)
 
bool LoadAllEvalData (const GenericVector< STRING > &filenames)
 
STRING RunEvalAsync (int iteration, const double *training_errors, const TessdataManager &model_mgr, int training_stage)
 
STRING RunEvalSync (int iteration, const double *training_errors, const TessdataManager &model_mgr, int training_stage, int verbosity)
 

Detailed Description

Definition at line 29 of file lstmtester.h.

Constructor & Destructor Documentation

◆ LSTMTester()

tesseract::LSTMTester::LSTMTester ( int64_t  max_memory)

Definition at line 26 of file lstmtester.cpp.

27  : test_data_(max_memory), total_pages_(0), async_running_(false) {}

Member Function Documentation

◆ LoadAllEvalData() [1/2]

bool tesseract::LSTMTester::LoadAllEvalData ( const STRING filenames_file)

Definition at line 32 of file lstmtester.cpp.

32  {
33  GenericVector<STRING> filenames;
34  if (!LoadFileLinesToStrings(filenames_file.c_str(), &filenames)) {
35  tprintf("Failed to load list of eval filenames from %s\n",
36  filenames_file.string());
37  return false;
38  }
39  return LoadAllEvalData(filenames);
40 }
DLLSYM void tprintf(const char *format,...)
Definition: tprintf.cpp:35
const char * c_str() const
Definition: strngs.cpp:205
const char * string() const
Definition: strngs.cpp:194
bool LoadFileLinesToStrings(const char *filename, GenericVector< STRING > *lines)
Definition: fileio.h:31
bool LoadAllEvalData(const STRING &filenames_file)
Definition: lstmtester.cpp:32

◆ LoadAllEvalData() [2/2]

bool tesseract::LSTMTester::LoadAllEvalData ( const GenericVector< STRING > &  filenames)

Definition at line 45 of file lstmtester.cpp.

45  {
46  test_data_.Clear();
47  bool result = test_data_.LoadDocuments(filenames, CS_SEQUENTIAL, nullptr);
48  total_pages_ = test_data_.TotalPages();
49  return result;
50 }
bool LoadDocuments(const GenericVector< STRING > &filenames, CachingStrategy cache_strategy, FileReader reader)
Definition: imagedata.cpp:580

◆ RunEvalAsync()

STRING tesseract::LSTMTester::RunEvalAsync ( int  iteration,
const double *  training_errors,
const TessdataManager model_mgr,
int  training_stage 
)

Definition at line 54 of file lstmtester.cpp.

56  {
57  STRING result;
58  if (total_pages_ == 0) {
59  result.add_str_int("No test data at iteration", iteration);
60  return result;
61  }
62  if (!LockIfNotRunning()) {
63  result.add_str_int("Previous test incomplete, skipping test at iteration",
64  iteration);
65  return result;
66  }
67  // Save the args.
68  STRING prev_result = test_result_;
69  test_result_ = "";
70  if (training_errors != nullptr) {
71  test_iteration_ = iteration;
72  test_training_errors_ = training_errors;
73  test_model_mgr_ = model_mgr;
74  test_training_stage_ = training_stage;
75  SVSync::StartThread(&LSTMTester::ThreadFunc, this);
76  } else {
77  UnlockRunning();
78  }
79  return prev_result;
80 }
static void StartThread(void *(*func)(void *), void *arg)
Create new thread.
Definition: svutil.cpp:81
void add_str_int(const char *str, int number)
Definition: strngs.cpp:377
Definition: strngs.h:45

◆ RunEvalSync()

STRING tesseract::LSTMTester::RunEvalSync ( int  iteration,
const double *  training_errors,
const TessdataManager model_mgr,
int  training_stage,
int  verbosity 
)

Definition at line 84 of file lstmtester.cpp.

86  {
87  LSTMTrainer trainer;
88  trainer.InitCharSet(model_mgr);
89  TFile fp;
90  if (!model_mgr.GetComponent(TESSDATA_LSTM, &fp) ||
91  !trainer.DeSerialize(&model_mgr, &fp)) {
92  return "Deserialize failed";
93  }
94  int eval_iteration = 0;
95  double char_error = 0.0;
96  double word_error = 0.0;
97  int error_count = 0;
98  while (error_count < total_pages_) {
99  const ImageData* trainingdata = test_data_.GetPageBySerial(eval_iteration);
100  trainer.SetIteration(++eval_iteration);
101  NetworkIO fwd_outputs, targets;
102  Trainability result =
103  trainer.PrepareForBackward(trainingdata, &fwd_outputs, &targets);
104  if (result != UNENCODABLE) {
105  char_error += trainer.NewSingleError(tesseract::ET_CHAR_ERROR);
106  word_error += trainer.NewSingleError(tesseract::ET_WORD_RECERR);
107  ++error_count;
108  if (verbosity > 1 || (verbosity > 0 && result != PERFECT)) {
109  tprintf("Truth:%s\n", trainingdata->transcription().string());
110  GenericVector<int> ocr_labels;
111  GenericVector<int> xcoords;
112  trainer.LabelsFromOutputs(fwd_outputs, &ocr_labels, &xcoords);
113  STRING ocr_text = trainer.DecodeLabels(ocr_labels);
114  tprintf("OCR :%s\n", ocr_text.string());
115  }
116  }
117  }
118  char_error *= 100.0 / total_pages_;
119  word_error *= 100.0 / total_pages_;
120  STRING result;
121  result.add_str_int("At iteration ", iteration);
122  result.add_str_int(", stage ", training_stage);
123  result.add_str_double(", Eval Char error rate=", char_error);
124  result.add_str_double(", Word error rate=", word_error);
125  return result;
126 }
DLLSYM void tprintf(const char *format,...)
Definition: tprintf.cpp:35
const ImageData * GetPageBySerial(int serial)
Definition: imagedata.h:344
const char * string() const
Definition: strngs.cpp:194
void add_str_int(const char *str, int number)
Definition: strngs.cpp:377
Definition: strngs.h:45
void add_str_double(const char *str, double number)
Definition: strngs.cpp:387

The documentation for this class was generated from the following files: