def load(instance, logger, dirname):
modulelist = load_modules(dirname)
logger.print_info("Loading {} modules".format(len(modulelist)))
for module in modulelist:
try:
modulelist[modulelist.index(module)] = module.Module(instance)
except BaseException as e:
continue
for module in modulelist:
if hasattr(module, "post_init"):
module.post_init(modulelist)
for method in dir(module):
if method[:3] == "on_":
try:
instance.connection.add_global_handler(
method[3:], getattr(module, method))
except Exception as e:
logger.print_error(
"Error {} occured while hooking module {}.".format(
e, module.info["name"]))
logger.print_info("Checking dependencies...")
for module in modulelist:
mods = [x.info["name"] for x in modulelist]
for depend in module.info["depends"]:
if depend not in mods:
logger.print_error(
"{0} depends on {1} but {1} does not exist! Download {1} or remove {0}."
.format(module.info["name"], depend))
logger.print_info("All dependencies are included.")
def convert(self, annot):
name = annot['filename']
if self.verbose:
print_info('Converting VIA annotations for {}'.format(name))
if not (self.input_dir / name).exists:
print_error('Original image {} not found'.format(name))
return None
size = Image.open(self.input_dir / name).size
img = Image.new(self.mode, size, color=self.background_color)
draw = ImageDraw.Draw(img)
for region in annot['regions']:
shape = region['shape_attributes']
if shape['name'] == 'circle':
cx, cy, r = shape['cx'], shape['cy'], shape['r']
bbox = [(cx - r, cy - r), (cx + r, cy + r)]
draw.ellipse(bbox, fill=self.color)
elif shape['name'] == 'ellipse':
cx, cy, rx, ry = shape['cx'], shape['cy'], shape['rx'], shape['ry']
bbox = [(cx - rx, cy - ry), (cx + rx, cy + ry)]
draw.ellipse(bbox, fill=self.color)
elif shape['name'] == 'polygon':
polygon = list(zip(shape['all_points_x'], shape['all_points_y']))
draw.polygon(polygon, fill=self.color)
else:
raise NotImplementedError('shape "{}" not implemented'.format(shape['name']))
return img
def get_resources_with_targets(manifest):
try:
canvases = manifest['sequences'][0]['canvases']
images_list = [canvas['images'] for canvas in canvases]
return [(image['resource']['@id'], image['on']) for images in images_list for image in images]
except KeyError as e:
print_error(e)
return []
def get_json(url):
try:
response = requests.get(url)
if response.ok:
return response.json()
else:
response.raise_for_status()
except requests.exceptions.RequestException as e:
print_error(e)
return None
def update_bot(): # we won't call this function from inside the bot
# Assuming you are inside the TextSiri git repository (you should be.)
try:
subprocess.check_output(["git", "fetch"])
subprocess.check_output(["git", "pull"])
except FileNotFoundError:
logger.print_error("You must have git installed!")
exit(1)
logger.print_info("Updated TextSiri, restarting.")
python = sys.executable
sys.argv.remove("--upgrade")
os.execl(python, python, *sys.argv)
def update_bot(): # we won't call this function from inside the bot
# Assuming you are inside the TextSiri git repository (you should be.)
try:
subprocess.check_output(["git", "fetch"])
subprocess.check_output(["git", "pull"])
except FileNotFoundError:
logger.print_error("You must have git installed!")
exit(1)
logger.print_info("Updated TextSiri, restarting.")
python = sys.executable
sys.argv.remove("--upgrade")
os.execl(python, python, * sys.argv)
def load_modules(dirname):
module_list = []
plugins_folder = [os.path.join(os.getcwd(), dirname)]
plugins = set(glob.glob(os.path.join(dirname, "*.py")))
for plugin in plugins:
if plugin.find("__") != -1:
continue
_plugin = os.path.join(os.getcwd(), plugin)
try:
moduleinfo = imp.find_module(
plugin.split(os.sep)[1].split(".")[0], plugins_folder)
module_object = imp.load_source(plugin, moduleinfo[1])
except ImportError as e:
if str(e).startswith('No module named'):
logger.print_error(
'Failed to load plugin {}: the plugin could not be found!'.
format(plugin))
else:
logger.print_error(
'Failed to load plugin {}: import error {}'.format(
plugin, str(e)))
except BaseException as e:
logger.print_error(
"The following error occured while importing module {}: \"{}\". Please fix the issue or contact the module author. "
.format(plugin, str(e)))
sys.exit(1)
else:
module_list.append(module_object)
return module_list
def message_loop(self, c):
sent_message = False
while True:
time.sleep(0.05)
if len(self.message_buffer) > 0:
if self.message_counter > 5:
self.message_slowmode_lock = 1
if self.message_slowmode_lock:
time.sleep(0.5)
if len(self.message_buffer) > 0:
command = self.message_buffer.pop(0)
command_name = command.pop(0)
try:
getattr(c, command_name)(command.pop(0), *command)
sent_message = True
except Exception as e:
logger.print_error(str(e))
if sent_message:
self.message_counter += 1
else:
self.message_counter -= 1
if self.message_counter == 0:
self.message_slowmode_lock = 0
def message_loop(self, c):
sent_message = False
while True:
time.sleep(0.05)
if len(self.message_buffer) > 0:
if self.message_counter > 5:
self.message_slowmode_lock = 1
if self.message_slowmode_lock:
time.sleep(0.5)
if len(self.message_buffer) > 0:
command = self.message_buffer.pop(0)
command_name = command.pop(0)
try:
getattr(c, command_name)(command.pop(0), *command)
sent_message = True
except Exception as e:
logger.print_error(str(e))
if sent_message:
self.message_counter += 1
else:
self.message_counter -= 1
if self.message_counter == 0:
self.message_slowmode_lock = 0
def load(instance, logger, dirname):
modulelist = load_modules(dirname)
logger.print_info("Loading {} modules".format(len(modulelist)))
for module in modulelist:
try:
modulelist[modulelist.index(module)] = module.Module(instance)
except BaseException as e:
continue
for module in modulelist:
if hasattr(module, "post_init"):
module.post_init(modulelist)
for method in dir(module):
if method[:3] == "on_":
try:
instance.connection.add_global_handler(method[3:], getattr(module, method))
except Exception as e:
logger.print_error("Error {} occured while hooking module {}.".format(e, module.info["name"]))
logger.print_info("Checking dependencies...")
for module in modulelist:
mods = [x.info["name"] for x in modulelist]
for depend in module.info["depends"]:
if depend not in mods:
logger.print_error("{0} depends on {1} but {1} does not exist! Download {1} or remove {0}.".format(module.info["name"], depend))
logger.print_info("All dependencies are included.")
def load_modules(dirname):
module_list = []
plugins_folder = [os.path.join(os.getcwd(), dirname)]
plugins = set(glob.glob(os.path.join(dirname, "*.py")))
for plugin in plugins:
if plugin.find("__") != -1:
continue
_plugin = os.path.join(os.getcwd(), plugin)
try:
moduleinfo = imp.find_module(plugin.split(os.sep)[1].split(".")[0], plugins_folder)
module_object = imp.load_source(plugin, moduleinfo[1])
except ImportError as e:
if str(e).startswith('No module named'):
logger.print_error('Failed to load plugin {}: the plugin could not be found!'.format(plugin))
else:
logger.print_error('Failed to load plugin {}: import error {}'.format(plugin, str(e)))
except BaseException as e:
logger.print_error("The following error occured while importing module {}: \"{}\". Please fix the issue or contact the module author. ".format(plugin, str(e)))
sys.exit(1)
else:
module_list.append(module_object)
return module_list
def print_and_log_error(self, string):
self.logger.error(string)
if self.verbose:
print_error(string)
logger.print_info(
"{color_green}New bot version found! Pass --upgrade to upgrade TextSiri at startup."
)
def update_bot(): # we won't call this function from inside the bot
# Assuming you are inside the TextSiri git repository (you should be.)
try:
subprocess.check_output(["git", "fetch"])
subprocess.check_output(["git", "pull"])
except FileNotFoundError:
logger.print_error("You must have git installed!")
exit(1)
logger.print_info("Updated TextSiri, restarting.")
python = sys.executable
sys.argv.remove("--upgrade")
os.execl(python, python, *sys.argv)
if __name__ == "__main__":
if "--upgrade" in sys.argv:
TextSiri.update_bot()
else:
Instance = TextSiri()
try:
Instance.start()
except KeyboardInterrupt:
Instance.die("Ctrl+C from terminal.")
sys.exit(0)
except Exception as e:
logger.print_error(type(e).__name__ + str(e))
def main():
parser = argparse.ArgumentParser(description="socr")
parser.add_argument('--name', type=str, default="dhSegment")
parser.add_argument('--lr',
type=float,
default=0.0001,
help="Learning rate")
parser.add_argument('--overlr',
action='store_const',
const=True,
default=False,
help="Override the learning rate")
parser.add_argument('--bs', type=int, default=16, help="The batch size")
parser.add_argument('--losstype',
type=str,
default='bce',
help="The loss type. Ex : mse, bce, norm")
parser.add_argument('--thicknesses',
type=int,
default=2,
help="Line thicknesses in the document")
parser.add_argument('--hystmin',
type=float,
default=0.5,
help="Hysteresys thresholding minimum")
parser.add_argument('--hystmax',
type=float,
default=0.5,
help="Hysteresys thresholding maximum")
parser.add_argument('--expdecay',
type=float,
default=0.98,
help="Exponential decay")
parser.add_argument(
'--heightimportance',
type=float,
default=0.001,
help="Height prediction importance during the training")
parser.add_argument('--weightdecay',
type=float,
default=0.000001,
help="Weight decay")
parser.add_argument('--epochlimit',
type=int,
default=None,
help="Limit the number of epoch")
parser.add_argument('--bnmomentum',
type=float,
default=0.1,
help="BatchNorm Momentum")
parser.add_argument('--disablecuda',
action='store_const',
const=True,
default=False,
help="Disable cuda")
parser.add_argument('--icdartrain',
type=str,
help="Path to the ICDAR Training set")
parser.add_argument('--icdartest',
type=str,
default=None,
help="Path to the ICDAR Testing set")
parser.add_argument('--generated',
action='store_const',
const=True,
default=False,
help="Enable generated data")
args = parser.parse_args()
model = dhSegment(args.losstype, args.hystmin, args.hystmax,
args.thicknesses, args.heightimportance, args.bnmomentum)
loss = model.create_loss()
if not args.disablecuda:
model = torch.nn.DataParallel(model.cuda())
loss = loss.cuda()
else:
model = CPUParallel(model.cpu())
loss = loss.cpu()
optimizer = torch.optim.Adam(model.parameters(),
lr=args.lr,
weight_decay=args.weightdecay)
adaptative_optimizer = torch.optim.lr_scheduler.ExponentialLR(
optimizer, args.expdecay)
os.makedirs('checkpoints', exist_ok=True)
checkpoint_name = "checkpoints/" + args.name + ".pth.tar"
epoch = 0
if os.path.exists(checkpoint_name):
print_normal("Restoring the weights...")
checkpoint = torch.load(checkpoint_name)
epoch = checkpoint['epoch']
model.load_state_dict(checkpoint['state_dict'])
optimizer.load_state_dict(checkpoint['optimizer'])
adaptative_optimizer.load_state_dict(
checkpoint['adaptative_optimizer'])
else:
print_warning("Can't find '" + checkpoint_name + "'")
if args.overlr is not None:
print_normal("Overwriting the lr to " + str(args.lr))
for param_group in optimizer.param_groups:
param_group['lr'] = args.lr
train_databases = [ICDARDocumentSet(args.icdartrain, loss, True)]
if args.generated:
sys.path.append(
os.path.join(os.path.dirname(os.path.abspath(__file__)),
"submodules/scribbler"))
from scribbler.generator import DocumentGenerator
train_databases.append(DocumentGenerator(loss))
train_database = torch.utils.data.ConcatDataset(train_databases)
test_database_path = None
if args.icdartest is not None:
test_database_path = args.icdartest
moving_average = MovingAverage(
max(train_database.__len__() // args.bs, 1024))
try:
while True:
if args.epochlimit is not None and epoch > args.epochlimit:
print_normal("Epoch " + str(args.epochlimit) + "reached !")
break
model.train()
loader = torch.utils.data.DataLoader(train_database,
batch_size=args.bs,
shuffle=True,
num_workers=4,
collate_fn=collate)
for i, data in enumerate(loader, 0):
inputs, labels = data
optimizer.zero_grad()
variable = torch.autograd.Variable(inputs).float()
labels = torch.autograd.Variable(labels).float()
if not args.disablecuda:
variable = variable.cuda()
labels = labels.cuda()
else:
variable = variable.cpu()
labels = labels.cpu()
outputs = model(variable)
loss_value = loss.forward(outputs, labels)
loss_value.backward()
loss_value_cpu = loss_value.data.cpu().numpy()
optimizer.step()
loss_value_np = float(loss_value.data.cpu().numpy())
moving_average.addn(loss_value_np)
if (i * args.bs) % 8 == 0:
sys.stdout.write(TerminalColors.BOLD + '[%d, %5d] ' %
(epoch + 1,
(i * args.bs) + 1) + TerminalColors.ENDC)
sys.stdout.write(
'lr: %.8f; loss: %.4f ; curr: %.4f ;\r' %
(optimizer.state_dict()['param_groups'][0]['lr'],
moving_average.moving_average(), loss_value_cpu))
epoch = epoch + 1
adaptative_optimizer.step()
sys.stdout.write("\n")
try:
if args.icdartest is not None:
callback(model, loss, test_database_path)
except Exception as e:
print_error("Can't test : " + str(e))
except KeyboardInterrupt:
pass
print_normal("Done training ! Saving...")
torch.save(
{
'epoch': epoch,
'state_dict': model.state_dict(),
'optimizer': optimizer.state_dict(),
'adaptative_optimizer': adaptative_optimizer.state_dict(),
}, checkpoint_name)
for i in range(1, len(local_version)):
if int(remote_version[i]) > int(local_version[i]):
logger.print_info("{color_green}New bot version found! Pass --upgrade to upgrade TextSiri at startup.")
def update_bot(): # we won't call this function from inside the bot
# Assuming you are inside the TextSiri git repository (you should be.)
try:
subprocess.check_output(["git", "fetch"])
subprocess.check_output(["git", "pull"])
except FileNotFoundError:
logger.print_error("You must have git installed!")
exit(1)
logger.print_info("Updated TextSiri, restarting.")
python = sys.executable
sys.argv.remove("--upgrade")
os.execl(python, python, * sys.argv)
if __name__ == "__main__":
if "--upgrade" in sys.argv:
TextSiri.update_bot()
else:
Instance = TextSiri()
try:
Instance.start()
except KeyboardInterrupt:
Instance.die("Ctrl+C from terminal.")
sys.exit(0)
except Exception as e:
logger.print_error(type(e).__name__+str(e))
