def setUpClass(cls):
export_config = get_config(config_file, section='export')
export_config['dataset'] = get_config(config_file, section='eval')['dataset']
cls.config = export_config
cls.config.update({'expected_outputs': expected_outputs})
cls.model_path = os.path.join(mkdtemp(), os.path.split(cls.config.get('model_path'))[1])
cls.res_model_name = os.path.join(os.path.dirname(cls.model_path), cls.config.get('res_model_name'))
cls.config['res_model_name'] = cls.res_model_name
cls.config['model_path'] = cls.model_path
if not os.path.exists(cls.model_path):
download_checkpoint(cls.model_path, cls.config.get('model_url'))
cls.exporter = Exporter(cls.config)
def setUpClass(cls):
train_config = get_config(config_file, section='train')
cls.config = train_config
cls.config['epochs'] = 1
cls.config['_test_steps'] = 40
cls.work_dir = mkdtemp()
cls.trainer = Trainer(work_dir=cls.work_dir, config=cls.config)
def setUpClass(cls):
test_config = get_config(config_file, section='eval')
cls.config = test_config
cls.config.update({'expected_outputs': expected_outputs})
if not os.path.exists(cls.config.get("model_path")):
download_checkpoint(cls.config.get("model_path"),
cls.config.get("model_url"))
cls.validator = Evaluator(config=cls.config)
def setUpClass(cls):
train_config = get_config(config_file, section='train')
cls.config = train_config
cls.config['epochs'] = 1
# workaround for training test without downloading language model (~4 Gb)
if cls.config['head'].get('use_semantics'):
cls.config['head']['use_semantics'] = False
cls.config['_test_steps'] = 40
cls.work_dir = mkdtemp()
cls.trainer = Trainer(work_dir=cls.work_dir, config=cls.config)
def setUpClass(cls):
train_config = get_config(config_file, section='train')
cls.config = train_config
cls.config['epochs'] = 1
# workaround for training test without downloading language model (~4 Gb)
if cls.config['head'].get('use_semantics'):
cls.config['head']['use_semantics'] = False
# workaround for training test without running it via `python -m torch.distributed.launch`
if cls.config.get('multi_gpu'):
cls.config['multi_gpu'] = False
cls.config['_test_steps'] = 40
cls.config['batch_size'] = 2 # only for this test
cls.work_dir = mkdtemp()
cls.trainer = Trainer(work_dir=cls.work_dir, config=cls.config)
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
"""
import argparse
import sys
from text_recognition.utils.get_config import get_config
from text_recognition.utils.trainer import Trainer
def parse_args():
args = argparse.ArgumentParser()
args.add_argument('--config')
args.add_argument('--work_dir')
return args.parse_args()
if __name__ == '__main__':
assert sys.version_info[0] == 3
arguments = parse_args()
train_config = get_config(arguments.config, section='train')
experiment = Trainer(work_dir=arguments.work_dir, config=train_config)
experiment.train()
pred = torch.nn.functional.log_softmax(logits.detach(), dim=2)
pred = ctc_greedy_search(pred, 0)
return self.vocab.construct_phrase(pred[0], ignore_end_token=self.use_ctc)
def parse_args():
args = argparse.ArgumentParser()
args.add_argument('--config')
args.add_argument('-i', '--input', help='Path to a folder with images or path to an image files', required=True)
return args.parse_args()
if __name__ == '__main__':
arguments = parse_args()
demo_config = get_config(arguments.config, section='demo')
demo = TextRecognitionDemo(demo_config)
try:
check_environment()
except EnvironmentError:
print('Warning: cannot render image because some render tools are not installed')
print('Check that pdflatex, ghostscript and imagemagick are installed')
print('For details, please, refer to README.md')
render_images = False
else:
render_images = True
if os.path.isdir(arguments.input):
inputs = sorted(os.path.join(arguments.input, inp)
for inp in os.listdir(arguments.input))
else:
inputs = [arguments.input]
import argparse
from text_recognition.utils.get_config import get_config
from text_recognition.utils.exporter import Exporter
def parse_args():
args = argparse.ArgumentParser()
args.add_argument('--config')
return args.parse_args()
if __name__ == '__main__':
arguments = parse_args()
export_config = get_config(arguments.config, section='export')
head_type = export_config.get('head').get('type')
exporter = Exporter(export_config)
if head_type == 'AttentionBasedLSTM':
exporter.export_encoder()
exporter.export_decoder()
elif head_type == 'LSTMEncoderDecoder':
exporter.export_complete_model()
print('Model succesfully exported to ONNX')
if export_config.get('export_ir'):
if head_type == 'AttentionBasedLSTM':
exporter.export_encoder_ir()
exporter.export_decoder_ir()
elif head_type == 'LSTMEncoderDecoder':
exporter.export_complete_model_ir()
print('Model succesfully exported to OpenVINO IR')
import argparse
import os
from text_recognition.utils.get_config import get_config
from text_recognition.utils.evaluator import Evaluator
def parse_args():
args = argparse.ArgumentParser()
args.add_argument('--config')
return args.parse_args()
if __name__ == '__main__':
arguments = parse_args()
test_config = get_config(arguments.config, section='eval')
validator = Evaluator(test_config)
if 'model_folder' in test_config.keys():
model_folder = test_config.get('model_folder')
best_model, best_result = None, 0
for model in os.listdir(model_folder):
validator.runner.reload_model(os.path.join(model_folder, model))
result = validator.validate()
if result > best_result:
best_result = result
best_model = os.path.join(model_folder, model)
print('model = {}'.format(best_model))
result = best_result
else:
result = validator.validate()
print('Result metric is: {}'.format(result))