def model_file_check():
if os.path.isfile(Config.IMAGE_MODEL_FILE):
return
Logger.info('正在下载模型文件...')
with requests.get(Config.OCR.get('image_model_url'), stream=True) as r:
with open(Config.IMAGE_MODEL_FILE, 'wb') as f:
shutil.copyfileobj(r.raw, f)
Logger.info('下载成功\n')
def model_file_check():
try:
if os.path.isfile(Config.IMAGE_MODEL_FILE):
model_hash_verify()
return
Logger.info('正在下载模型文件...')
with requests.get(Config.OCR.get('image_model_url'), stream=True) as r:
with open(Config.IMAGE_MODEL_FILE, 'wb') as f:
shutil.copyfileobj(r.raw, f)
Logger.info('下载成功\n')
model_hash_verify()
except InvalidModelException:
return model_file_check()
def get_coordinate(self, img_str):
# 储存最终坐标结果
result = ''
try:
# 读取并预处理验证码
img = cv2.imdecode(np.fromstring(img_str, np.uint8),
cv2.IMREAD_COLOR)
text = self.get_text(img)
images = np.array(list(self._get_imgs(img)))
images = self.preprocess_input(images)
label = self.model.predict(text)
label = label.argmax()
text = self.texts[label]
# list放文字
titles = [text]
position = []
# 获取下一个词
# 根据第一个词的长度来定位第二个词的位置
if len(text) == 1:
offset = 27
elif len(text) == 2:
offset = 47
else:
offset = 60
text2 = self.get_text(img, offset=offset)
if text2.mean() < 0.95:
label = self.model.predict(text2)
label = label.argmax()
text2 = self.texts[label]
titles.append(text2)
labels = self.code_model.predict(images)
labels = labels.argmax(axis=1)
for pos, label in enumerate(labels):
if self.texts[label] in titles:
position.append(pos + 1)
# 没有识别到结果
if len(position) == 0:
return result
result = position
Logger.info('识别结果: %s' % result)
except:
pass
return result
class IpUtils:
def __init__(self, config: Config, logger: Logger):
self._config = Config() if config is None else config
self._logger = Logger() if logger is None else logger
def get_external_ip(self) -> str:
external_ip = requests.get(self._config.get_ip_provider_url()).text
self._logger.info('External ip address: %s' % external_ip)
return external_ip
@staticmethod
def gather_ip_from_dns_response(dns_response_content: str) -> str:
match = re.search(r'\"data\":\"(\d{1,4}\.\d{1,4}\.\d{1,4}\.\d{1,4})\"',
dns_response_content)
if match:
return match.group(1)
else:
raise ValueError('Ip cannot be gathered from dns response: %s' %
dns_response_content)
dropout, device, hid_act, pooler_act, add_pooler
"""
device = 'cuda:0' if cfg.config['cuda'] else 'cpu'
bert_checkpoint = torch.load('saved_models/biobert-base-cased-v1.1.pth')
dense_layer_checkpoint = torch.load('saved_models/to_embedding.pt')
n_classes = len(train_data_loader.dataset.label2id)
model = MultiInstanceLearning(n_classes=n_classes,
vocab_size=len(vocab),
device=device,
bert_checkpoint=bert_checkpoint,
dense_layer_checkpoint=dense_layer_checkpoint,
**cfg.config)
model.to(device)
logger.info(model)
# optimizer and criterion
param = filter(lambda p: p.requires_grad, model.parameters())
# param = [p for p in model.parameters() if p.requires_grad]
optimizer = get_optimizer(cfg.config['optimizer'], param, lr=cfg.config['lr'])
lr_scheduler = ReduceLROnPlateau(optimizer,
'max',
factor=cfg.config['lr_decay'],
patience=cfg.config['lr_decay_patience'])
criterion = nn.CrossEntropyLoss(
weight=train_data_loader.dataset.weights.to(device))
#trainer
trainer = MILTrainer(model=model,
DB_INFO_ALERT_SETTINGS)
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--load', action='store_const', const=True)
parser.add_argument('--save', action='store_const', const=True)
parser.add_argument('--drop_elastic', action='store_const', const=True)
parser.add_argument('--migrate', action='store_const', const=True)
options = parser.parse_args(sys.argv[1:])
col = Collector(base_path=BASE_PATH, sources=DATA_SOURCES)
if options.load:
logger.info('Action --load triggered', alert=True)
col.load_all() # TODO добавить какой-нибудь report об окончании загрузки с информацией о результатах
logger.info('Action --load completed', alert=True)
if options.save:
logger.info('Action --save triggered', alert=True)
report = col.save_all()
logger.info(report, alert=True)
logger.info('Action --save completed', alert=True)
if options.drop_elastic: