// Sets the mapped_features_ from the features_ using the provided
// feature_space to the indexed versions of the features.
void TrainingSample::IndexFeatures(const IntFeatureSpace& feature_space) {
GenericVector<int> indexed_features;
feature_space.IndexAndSortFeatures(features_, num_features_,
&mapped_features_);
features_are_indexed_ = true;
features_are_mapped_ = false;
}
// Display the samples with the given indexed feature that also match
// the given shape.
void TrainingSampleSet::DisplaySamplesWithFeature(int f_index,
const Shape &shape,
const IntFeatureSpace &space,
ScrollView::Color color,
ScrollView *window) const {
for (int s = 0; s < num_raw_samples(); ++s) {
const TrainingSample *sample = GetSample(s);
if (shape.ContainsUnichar(sample->class_id())) {
GenericVector <int> indexed_features;
space.IndexAndSortFeatures(sample->features(), sample->num_features(),
&indexed_features);
for (int f = 0; f < indexed_features.size(); ++f) {
if (indexed_features[f] == f_index) {
sample->DisplayFeatures(color, window);
}
}
}
}
}
/**
* Creates a MasterTraininer and loads the training data into it:
* Initializes feature_defs and IntegerFX.
* Loads the shape_table if shape_table != nullptr.
* Loads initial unicharset from -U command-line option.
* If FLAGS_T is set, loads the majority of data from there, else:
* - Loads font info from -F option.
* - Loads xheights from -X option.
* - Loads samples from .tr files in remaining command-line args.
* - Deletes outliers and computes canonical samples.
* - If FLAGS_output_trainer is set, saves the trainer for future use.
* Computes canonical and cloud features.
* If shape_table is not nullptr, but failed to load, make a fake flat one,
* as shape clustering was not run.
*/
MasterTrainer* LoadTrainingData(int argc, const char* const * argv,
bool replication,
ShapeTable** shape_table,
STRING* file_prefix) {
InitFeatureDefs(&feature_defs);
InitIntegerFX();
*file_prefix = "";
if (!FLAGS_D.empty()) {
*file_prefix += FLAGS_D.c_str();
*file_prefix += "/";
}
// If we are shape clustering (nullptr shape_table) or we successfully load
// a shape_table written by a previous shape clustering, then
// shape_analysis will be true, meaning that the MasterTrainer will replace
// some members of the unicharset with their fragments.
bool shape_analysis = false;
if (shape_table != nullptr) {
*shape_table = LoadShapeTable(*file_prefix);
if (*shape_table != nullptr) shape_analysis = true;
} else {
shape_analysis = true;
}
MasterTrainer* trainer = new MasterTrainer(NM_CHAR_ANISOTROPIC,
shape_analysis,
replication,
FLAGS_debug_level);
IntFeatureSpace fs;
fs.Init(kBoostXYBuckets, kBoostXYBuckets, kBoostDirBuckets);
if (FLAGS_T.empty()) {
trainer->LoadUnicharset(FLAGS_U.c_str());
// Get basic font information from font_properties.
if (!FLAGS_F.empty()) {
if (!trainer->LoadFontInfo(FLAGS_F.c_str())) {
delete trainer;
return nullptr;
}
}
if (!FLAGS_X.empty()) {
if (!trainer->LoadXHeights(FLAGS_X.c_str())) {
delete trainer;
return nullptr;
}
}
trainer->SetFeatureSpace(fs);
const char* page_name;
// Load training data from .tr files on the command line.
while ((page_name = GetNextFilename(argc, argv)) != nullptr) {
tprintf("Reading %s ...\n", page_name);
trainer->ReadTrainingSamples(page_name, feature_defs, false);
// If there is a file with [lang].[fontname].exp[num].fontinfo present,
// read font spacing information in to fontinfo_table.
int pagename_len = strlen(page_name);
char *fontinfo_file_name = new char[pagename_len + 7];
strncpy(fontinfo_file_name, page_name, pagename_len - 2); // remove "tr"
strcpy(fontinfo_file_name + pagename_len - 2, "fontinfo"); // +"fontinfo"
trainer->AddSpacingInfo(fontinfo_file_name);
delete[] fontinfo_file_name;
// Load the images into memory if required by the classifier.
if (FLAGS_load_images) {
STRING image_name = page_name;
// Chop off the tr and replace with tif. Extension must be tif!
image_name.truncate_at(image_name.length() - 2);
image_name += "tif";
trainer->LoadPageImages(image_name.string());
}
}
trainer->PostLoadCleanup();
// Write the master trainer if required.
if (!FLAGS_output_trainer.empty()) {
FILE* fp = fopen(FLAGS_output_trainer.c_str(), "wb");
if (fp == nullptr) {
tprintf("Can't create saved trainer data!\n");
} else {
trainer->Serialize(fp);
fclose(fp);
}
}
} else {
bool success = false;
tprintf("Loading master trainer from file:%s\n",
FLAGS_T.c_str());
FILE* fp = fopen(FLAGS_T.c_str(), "rb");
if (fp == nullptr) {
tprintf("Can't read file %s to initialize master trainer\n",
FLAGS_T.c_str());
} else {
success = trainer->DeSerialize(false, fp);
fclose(fp);
}
if (!success) {
tprintf("Deserialize of master trainer failed!\n");
delete trainer;
return nullptr;
}
trainer->SetFeatureSpace(fs);
}
trainer->PreTrainingSetup();
if (!FLAGS_O.empty() &&
!trainer->unicharset().save_to_file(FLAGS_O.c_str())) {
fprintf(stderr, "Failed to save unicharset to file %s\n", FLAGS_O.c_str());
delete trainer;
return nullptr;
}
if (shape_table != nullptr) {
// If we previously failed to load a shapetable, then shape clustering
// wasn't run so make a flat one now.
if (*shape_table == nullptr) {
*shape_table = new ShapeTable;
trainer->SetupFlatShapeTable(*shape_table);
tprintf("Flat shape table summary: %s\n",
(*shape_table)->SummaryStr().string());
}
(*shape_table)->set_unicharset(trainer->unicharset());
}
return trainer;
}
// Delete outlier samples with few features that are shared with others.
// IndexFeatures must have been called already.
void TrainingSampleSet::DeleteOutliers(const IntFeatureSpace &feature_space,
bool debug) {
if (font_class_array_ == NULL)
OrganizeByFontAndClass();
Pixa *pixa = NULL;
if (debug)
pixa = pixaCreate(0);
GenericVector <int> feature_counts;
int fs_size = feature_space.Size();
int font_size = font_id_map_.CompactSize();
for (int font_index = 0; font_index < font_size; ++font_index) {
for (int c = 0; c < unicharset_size_; ++c) {
// Create a histogram of the features used by all samples of this
// font/class combination.
feature_counts.init_to_size(fs_size, 0);
FontClassInfo &fcinfo = (*font_class_array_)(font_index, c);
int sample_count = fcinfo.samples.size();
if (sample_count < kMinOutlierSamples)
continue;
for (int i = 0; i < sample_count; ++i) {
int s = fcinfo.samples[i];
const GenericVector <int> &features = samples_[s]->indexed_features();
for (int f = 0; f < features.size(); ++f) {
++feature_counts[features[f]];
}
}
for (int i = 0; i < sample_count; ++i) {
int s = fcinfo.samples[i];
const TrainingSample &sample = *samples_[s];
const GenericVector <int> &features = sample.indexed_features();
// A feature that has a histogram count of 1 is only used by this
// sample, making it 'bad'. All others are 'good'.
int good_features = 0;
int bad_features = 0;
for (int f = 0; f < features.size(); ++f) {
if (feature_counts[features[f]] > 1)
++good_features;
else
++bad_features;
}
// If more than 1/3 features are bad, then this is an outlier.
if (bad_features * 2 > good_features) {
tprintf("Deleting outlier sample of %s, %d good, %d bad\n",
SampleToString(sample).string(),
good_features, bad_features);
if (debug) {
pixaAddPix(pixa, sample.RenderToPix(&unicharset_), L_INSERT);
// Add the previous sample as well, so it is easier to see in
// the output what is wrong with this sample.
int t;
if (i == 0)
t = fcinfo.samples[1];
else
t = fcinfo.samples[i - 1];
const TrainingSample &csample = *samples_[t];
pixaAddPix(pixa, csample.RenderToPix(&unicharset_), L_INSERT);
}
// Mark the sample for deletion.
KillSample(samples_[s]);
}
}
}
}
// Truly delete all bad samples and renumber everything.
DeleteDeadSamples();
if (pixa != NULL) {
Pix *pix = pixaDisplayTiledInRows(pixa, 1, 2600, 1.0, 0, 10, 10);
pixaDestroy(&pixa);
pixWrite("outliers.png", pix, IFF_PNG);
pixDestroy(&pix);
}
}