def _configure_executor(self):
self.executor = Executor(self.algorithm, False, self.globalConfig)
self.executor.configure(
softplacement=not self.globalConfig["hardplacement"],
logplacement=self.globalConfig["logplacement"],
device=self.globalConfig["gpu"])
self.executor.restore(
os.path.join(self.config["model_path"],
"model-{}".format(self.config["model_epoch"])))
self.separator = SeparationRunner(config=self.modelConfig,
dataset=self.dataset,
subset="")
self.executables = [self.separator]
def _configure_executor(self):
self.executor = Executor(self.algorithm, False, self.globalConfig)
self.executor.configure(
softplacement=not self.globalConfig["hardplacement"],
logplacement=self.globalConfig["logplacement"],
device=self.globalConfig["gpu"])
self.executor.restore(
os.path.join(self.config["model_path"],
"model-{}".format(self.config["model_epoch"])))
if self.config["classify"]:
self.transcriber = RecClassRunner(self.dataset,
config=self.modelConfig)
else:
self.transcriber = RecognitionRunner(self.dataset,
config=self.modelConfig)
self.executables = [self.transcriber]
class TextSeparation(object):
name = "Text Separation"
def __init__(self, globalConfig={}, config={}):
self.globalConfig = Configuration(globalConfig, GLOBAL_DEFAULTS)
self.config = Configuration(config, DEFAULTS)
self.modelConfig = Configuration.load(self.config["model_path"],
"algorithm")
self._configure_dataset()
self._configure_algorithm()
self._configure_executor()
def _configure_algorithm(self):
self.algorithm = TFUnet(self.modelConfig["algo_config"])
self.algorithm.set_cpu(self.globalConfig['gpu'] == -1)
self.algorithm.configure(
slice_width=self.modelConfig['data_config.slice_width'],
slice_height=self.modelConfig['data_config.slice_height'])
def _configure_dataset(self):
self.dataset = PaperNoteSlicesSingle(
slice_width=self.modelConfig['data_config.slice_width'],
slice_height=self.modelConfig['data_config.slice_height'],
binarize=self.modelConfig.default("binary", False),
binarize_method=self.config["binarize_method"])
def _configure_executor(self):
self.executor = Executor(self.algorithm, False, self.globalConfig)
self.executor.configure(
softplacement=not self.globalConfig["hardplacement"],
logplacement=self.globalConfig["logplacement"],
device=self.globalConfig["gpu"])
self.executor.restore(
os.path.join(self.config["model_path"],
"model-{}".format(self.config["model_epoch"])))
self.separator = SeparationRunner(config=self.modelConfig,
dataset=self.dataset,
subset="")
self.executables = [self.separator]
def __call__(self, image, file):
original = self.dataset.set_image(image).copy()
self.executor(self.executables, auto_close=False)
outputs = np.argmax(np.asarray(self.separator.outputs), 3)
merged = self.dataset.merge_slices(outputs, original.shape)
output = (255 - (1 - merged) * (255 - original))
return np.uint8(output)
def close(self):
self.executor.close()
logger = Logger()
config = Configuration.load(model_folder, "algorithm")
algorithm = HtrNet(config['algo_config'])
dataset = PreparedDataset.PreparedDataset(config['dataset'], False,
config['data_config'])
algorithm.configure(batch_size=config['batch'],
learning_rate=config['learning_rate'],
sequence_length=dataset.max_length,
image_height=dataset.meta["height"],
image_width=dataset.meta["width"],
vocab_length=dataset.vocab_length,
channels=dataset.channels,
class_learning_rate=config.default(
'class_learning_rate', config['learning_rate']))
executor = Executor(algorithm, True, config, logger=logger)
paper_note_dataset = PaperNoteWords(paper_note_path=args.paper_note_path,
meta=dataset.meta,
vocab=dataset.vocab,
data_config=dataset.data_config,
max_length=dataset.max_length)
executor.configure(softplacement=not args.hardplacement,
logplacement=args.logplacement,
device=args.gpu)
executor.restore(models_path)
executables = [
RecognitionValidator(logger=logger,
config=config,
def __call__(self, subcommand, path):
command = [self.GIT]
return Executor(command) \
.with_cwd(path) \
.with_subcommand(subcommand)
class TranscriptionAndClassification(object):
def __init__(self, globalConfig={}, config={}):
self.globalConfig = Configuration(globalConfig, GLOBAL_DEFAULTS)
self.config = Configuration(config, DEFAULTS)
self.modelConfig = Configuration.load(self.config["model_path"],
"algorithm")
self._configure_dataset()
self._configure_algorithm()
self._configure_executor()
def _configure_algorithm(self):
self.algorithm = HtrNet(self.modelConfig["algo_config"])
self.algorithm.set_cpu(self.globalConfig['gpu'] == -1)
self.algorithm.configure(
batch_size=self.modelConfig['batch'],
learning_rate=self.modelConfig['learning_rate'],
sequence_length=self.dataset.max_length,
image_height=self.dataset.meta["height"],
image_width=self.dataset.meta["width"],
vocab_length=self.dataset.vocab_length,
channels=self.dataset.channels,
class_learning_rate=self.modelConfig.default(
'class_learning_rate', self.modelConfig['learning_rate']))
def _configure_dataset(self):
self.dataset = RegionDataset(
None,
self.config["model_path"],
data_config=self.modelConfig["data_config"])
self.dataset.scaling(self.config["scaling"], self.config["max_height"],
self.config["max_width"])
def _configure_executor(self):
self.executor = Executor(self.algorithm, False, self.globalConfig)
self.executor.configure(
softplacement=not self.globalConfig["hardplacement"],
logplacement=self.globalConfig["logplacement"],
device=self.globalConfig["gpu"])
self.executor.restore(
os.path.join(self.config["model_path"],
"model-{}".format(self.config["model_epoch"])))
if self.config["classify"]:
self.transcriber = RecClassRunner(self.dataset,
config=self.modelConfig)
else:
self.transcriber = RecognitionRunner(self.dataset,
config=self.modelConfig)
self.executables = [self.transcriber]
def __call__(self, images, file):
self.dataset.set_regions(images)
self.executor(self.executables, auto_close=False)
for idx in range(len(self.transcriber.transcriptions['trans'])):
text = self.transcriber.transcriptions['trans'][idx]
images[idx].set_text(text, self.dataset.decompile(text))
if self.config["classify"]:
score = self.transcriber.transcriptions['class'][idx]
images[idx].set_class(score,
score > self.config["class_thresh"])
return images
def close(self):
self.executor.close()