def load_model(config_file, device):
config_folder = os.path.join('config_folder', config_file)
option_file = os.path.join(config_folder, 'options.pickle')
checkpoint_file = os.path.join(config_folder, 'checkpoint.pyt')
checkpoint = torch.load(checkpoint_file, map_location='cpu')
_, hidden_config, noise_config = utils.load_options(option_file)
noiser = Noiser(noise_config, device)
model = Hidden(hidden_config, device, noiser, tb_logger=None)
utils.model_from_checkpoint(model, checkpoint)
return model
def init_prepare(args):
args.device = torch.device("cuda" if not args.disable_gpu
and torch.cuda.is_available() else "cpu")
args.output = f'{args.run_folder}/output'
utils.ensure_dir(args.output)
if args.arch == 'hidden':
args.options_file = f'{args.run_folder}/options-and-config.pickle'
args.checkpoint_file = f'{args.run_folder}/checkpoints/hidden-best-model.pyt'
train_options, hidden_config, noise_config = utils.load_options(
args.options_file)
noiser = Noiser(noise_config, device=args.device)
checkpoint = torch.load(args.checkpoint_file, map_location=args.device)
hidden_net = Hidden(hidden_config, args.device, noiser, None)
utils.model_from_checkpoint(hidden_net, checkpoint)
args.model = hidden_net
args.block_size = hidden_config.H
args.message_length = hidden_config.message_length
args.hidden_config = hidden_config
args.in_channels = hidden_config.input_channels
elif args.arch == 'ms-hidden':
checkpoint = torch.load(f'{args.run_folder}/trained-model.pth')
options = argparse.Namespace(**checkpoint['option'])
options.device = args.device
noiser = Noiser(options.noise, device=args.device)
model = nets.MS_Hidden(options, noiser).to(args.device)
model.load_state_dict(checkpoint['model'])
args.model = model
args.block_size = options.block_size
args.message_length = options.message
args.in_channels = options.in_channels
args.layers = options.layers
def train(model: Hidden, device: torch.device,
hidden_config: HiDDenConfiguration, train_options: TrainingOptions,
this_run_folder: str, tb_logger):
"""
Trains the HiDDeN model
:param model: The model
:param device: torch.device object, usually this is GPU (if avaliable), otherwise CPU.
:param hidden_config: The network configuration
:param train_options: The training settings
:param this_run_folder: The parent folder for the current training run to store training artifacts/results/logs.
:param tb_logger: TensorBoardLogger object which is a thin wrapper for TensorboardX logger.
Pass None to disable TensorboardX logging
:return:
"""
train_data, val_data = utils.get_data_loaders(hidden_config, train_options)
images_to_save = 8
saved_images_size = (512, 512)
best_epoch = train_options.best_epoch
best_cond = train_options.best_cond
for epoch in range(train_options.start_epoch,
train_options.number_of_epochs + 1):
logging.info(
f'\nStarting epoch {epoch}/{train_options.number_of_epochs} [{best_epoch}]'
)
training_losses = defaultdict(functions.AverageMeter)
epoch_start = time.time()
for image, _ in tqdm(train_data, ncols=80):
image = image.to(device) #.squeeze(0)
message = torch.Tensor(
np.random.choice(
[0, 1],
(image.shape[0], hidden_config.message_length))).to(device)
losses, _ = model.train_on_batch([image, message])
for name, loss in losses.items():
training_losses[name].update(loss)
train_duration = time.time() - epoch_start
logging.info('Epoch {} training duration {:.2f} sec'.format(
epoch, train_duration))
logging.info('-' * 40)
utils.write_losses(os.path.join(this_run_folder, 'train.csv'),
training_losses, epoch, train_duration)
if tb_logger is not None:
tb_logger.save_losses('train_loss', training_losses, epoch)
tb_logger.save_grads(epoch)
tb_logger.save_tensors(epoch)
tb_logger.writer.flush()
validation_losses = defaultdict(functions.AverageMeter)
logging.info('Running validation for epoch {}/{}'.format(
epoch, train_options.number_of_epochs))
val_image_patches = ()
val_encoded_patches = ()
val_noised_patches = ()
for image, _ in tqdm(val_data, ncols=80):
image = image.to(device) #.squeeze(0)
message = torch.Tensor(
np.random.choice(
[0, 1],
(image.shape[0], hidden_config.message_length))).to(device)
losses, (encoded_images, noised_images,
decoded_messages) = model.validate_on_batch(
[image, message])
for name, loss in losses.items():
validation_losses[name].update(loss)
if hidden_config.enable_fp16:
image = image.float()
encoded_images = encoded_images.float()
pick = np.random.randint(0, image.shape[0])
val_image_patches += (F.interpolate(
image[pick:pick + 1, :, :, :].cpu(),
size=(hidden_config.W, hidden_config.H)), )
val_encoded_patches += (F.interpolate(
encoded_images[pick:pick + 1, :, :, :].cpu(),
size=(hidden_config.W, hidden_config.H)), )
val_noised_patches += (F.interpolate(
noised_images[pick:pick + 1, :, :, :].cpu(),
size=(hidden_config.W, hidden_config.H)), )
if tb_logger is not None:
tb_logger.save_losses('val_loss', validation_losses, epoch)
tb_logger.writer.flush()
val_image_patches = torch.stack(val_image_patches).squeeze(1)
val_encoded_patches = torch.stack(val_encoded_patches).squeeze(1)
val_noised_patches = torch.stack(val_noised_patches).squeeze(1)
utils.save_images(val_image_patches[:images_to_save, :, :, :],
val_encoded_patches[:images_to_save, :, :, :],
val_noised_patches[:images_to_save, :, :, :],
epoch,
os.path.join(this_run_folder, 'images'),
resize_to=saved_images_size)
curr_cond = validation_losses['encoder_mse'].avg + validation_losses[
'bitwise-error'].avg
if best_cond is None or curr_cond < best_cond:
best_cond = curr_cond
best_epoch = epoch
utils.log_progress(validation_losses)
logging.info('-' * 40)
utils.save_checkpoint(model, train_options.experiment_name, epoch,
best_epoch, best_cond,
os.path.join(this_run_folder, 'checkpoints'))
logging.info(
f'Current best epoch = {best_epoch}, loss = {best_cond:.6f}')
utils.write_losses(os.path.join(this_run_folder, 'validation.csv'),
validation_losses, epoch,
time.time() - epoch_start)
def main():
# device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
device = torch.device(
'cuda') if torch.cuda.is_available() else torch.device('cpu')
parser = argparse.ArgumentParser(description='Training of HiDDeN nets')
parser.add_argument('--hostname',
default=socket.gethostname(),
help='the host name of the running server')
# parser.add_argument('--size', '-s', default=128, type=int, help='The size of the images (images are square so this is height and width).')
parser.add_argument('--data-dir',
'-d',
required=True,
type=str,
help='The directory where the data is stored.')
parser.add_argument(
'--runs_root',
'-r',
default=os.path.join('.', 'experiments'),
type=str,
help='The root folder where data about experiments are stored.')
parser.add_argument('--batch-size',
'-b',
default=1,
type=int,
help='Validation batch size.')
args = parser.parse_args()
if args.hostname == 'ee898-System-Product-Name':
args.data_dir = '/home/ee898/Desktop/chaoning/ImageNet'
args.hostname = 'ee898'
elif args.hostname == 'DL178':
args.data_dir = '/media/user/SSD1TB-2/ImageNet'
else:
args.data_dir = '/workspace/data_local/imagenet_pytorch'
assert args.data_dir
print_each = 25
completed_runs = [
o for o in os.listdir(args.runs_root)
if os.path.isdir(os.path.join(args.runs_root, o))
and o != 'no-noise-defaults'
]
print(completed_runs)
write_csv_header = True
current_run = args.runs_root
print(f'Run folder: {current_run}')
options_file = os.path.join(current_run, 'options-and-config.pickle')
train_options, hidden_config, noise_config = utils.load_options(
options_file)
train_options.train_folder = os.path.join(args.data_dir, 'val')
train_options.validation_folder = os.path.join(args.data_dir, 'val')
train_options.batch_size = args.batch_size
checkpoint, chpt_file_name = utils.load_last_checkpoint(
os.path.join(current_run, 'checkpoints'))
print(f'Loaded checkpoint from file {chpt_file_name}')
noiser = Noiser(noise_config, device, 'jpeg')
model = Hidden(hidden_config, device, noiser, tb_logger=None)
utils.model_from_checkpoint(model, checkpoint)
print('Model loaded successfully. Starting validation run...')
_, val_data = utils.get_data_loaders(hidden_config, train_options)
file_count = len(val_data.dataset)
if file_count % train_options.batch_size == 0:
steps_in_epoch = file_count // train_options.batch_size
else:
steps_in_epoch = file_count // train_options.batch_size + 1
with torch.no_grad():
noises = ['webp_10', 'webp_25', 'webp_50', 'webp_75', 'webp_90']
for noise in noises:
losses_accu = {}
step = 0
for image, _ in val_data:
step += 1
image = image.to(device)
message = torch.Tensor(
np.random.choice(
[0, 1], (image.shape[0],
hidden_config.message_length))).to(device)
losses, (
encoded_images, noised_images,
decoded_messages) = model.validate_on_batch_specific_noise(
[image, message], noise=noise)
if not losses_accu: # dict is empty, initialize
for name in losses:
losses_accu[name] = AverageMeter()
for name, loss in losses.items():
losses_accu[name].update(loss)
if step % print_each == 0 or step == steps_in_epoch:
print(f'Step {step}/{steps_in_epoch}')
utils.print_progress(losses_accu)
print('-' * 40)
# utils.print_progress(losses_accu)
write_validation_loss(os.path.join(args.runs_root,
'validation_run.csv'),
losses_accu,
noise,
checkpoint['epoch'],
write_header=write_csv_header)
write_csv_header = False
def train(model: Hidden, device: torch.device,
hidden_config: HiDDenConfiguration, train_options: TrainingOptions,
this_run_folder: str, tb_logger):
"""
Trains the HiDDeN model
:param model: The model
:param device: torch.device object, usually this is GPU (if avaliable), otherwise CPU.
:param hidden_config: The network configuration
:param train_options: The training settings
:param this_run_folder: The parent folder for the current training run to store training artifacts/results/logs.
:param tb_logger: TensorBoardLogger object which is a thin wrapper for TensorboardX logger.
Pass None to disable TensorboardX logging
:return:
"""
train_data, val_data = utils.get_data_loaders(hidden_config, train_options)
file_count = len(train_data.dataset)
if file_count % train_options.batch_size == 0:
steps_in_epoch = file_count // train_options.batch_size
else:
steps_in_epoch = file_count // train_options.batch_size + 1
print_each = 10
images_to_save = 8
saved_images_size = (512, 512)
for epoch in range(train_options.start_epoch,
train_options.number_of_epochs + 1):
logging.info('\nStarting epoch {}/{}'.format(
epoch, train_options.number_of_epochs))
logging.info('Batch size = {}\nSteps in epoch = {}'.format(
train_options.batch_size, steps_in_epoch))
training_losses = defaultdict(AverageMeter)
epoch_start = time.time()
step = 1
#train
for image, _ in train_data:
image = image.to(device)
"""
message = torch.Tensor(np.random.choice([0, 1], (image.shape[0], hidden_config.message_length))).to(device)
losses, _ = model.train_on_batch([image, message])
print(losses)
"""
#crop imgs
imgs = cropImg(32, image)
#iterate img
bitwise_arr = []
main_losses = None
encoded_imgs = []
for img in imgs:
img = img.to(device)
message = torch.Tensor(
np.random.choice(
[0, 1], (img.shape[0],
hidden_config.message_length))).to(device)
losses, (encoded_images, noised_images,
decoded_messages) = model.train_on_batch(
[img, message])
encoded_imgs.append(
encoded_images[0][0].cpu().detach().numpy())
main_losses = losses
for name, loss in losses.items():
if (name == 'bitwise-error '):
bitwise_arr.append(loss)
Total = 0
Vcount = 0
V_average = 0
H_average = 0
for i in range(0, len(encoded_imgs) - 1):
if ((i + 1) % 4 != 0):
img = encoded_imgs[i]
img_next = encoded_imgs[i + 1]
average_img = 0
average_img_next = 0
for j in range(0, 32):
for k in range(0, 10):
average_img = average_img + img[j][31 - k]
average_img_next = average_img_next + img_next[j][k]
average_blocking = np.abs(average_img -
average_img_next) / 320
V_average = V_average + average_blocking
for j in range(0, 32):
distinct = np.abs(img[j][31] - img_next[j][0])
Total = Total + 1
if (distinct > 0.5):
Vcount = Vcount + 1
V_average = V_average / 12
Hcount = 0
for i in range(0, len(encoded_imgs) - 4):
img = encoded_imgs[i]
img_next = encoded_imgs[i + 4]
average_img = 0
average_img_next = 0
for j in range(0, 32):
for k in range(0, 10):
average_img = average_img + img[31 - k][j]
average_img_next = average_img_next + img_next[k][j]
average_blocking = np.abs(average_img - average_img_next) / 320
H_average = H_average + average_blocking
for j in range(0, 32):
distinct = np.abs(img[31][j] - img_next[0][j])
Total = Total + 1
if (distinct > 0.5):
Hcount = Hcount + 1
H_average = H_average / 12
bitwise_arr = np.array(bitwise_arr)
bitwise_avg = np.average(bitwise_arr)
#blocking_loss = (Vcount+Hcount)/Total
blocking_loss = (H_average + V_average) / 2
for name, loss in main_losses.items():
if (name == 'bitwise-error '):
training_losses[name].update(bitwise_avg)
else:
if (name == 'blocking_effect'):
training_losses[name].update(blocking_loss)
else:
training_losses[name].update(loss)
if step % print_each == 0 or step == steps_in_epoch:
logging.info('Epoch: {}/{} Step: {}/{}'.format(
epoch, train_options.number_of_epochs, step,
steps_in_epoch))
utils.log_progress(training_losses)
logging.info('-' * 40)
step += 1
train_duration = time.time() - epoch_start
logging.info('Epoch {} training duration {:.2f} sec'.format(
epoch, train_duration))
logging.info('-' * 40)
utils.write_losses(os.path.join(this_run_folder, 'train.csv'),
training_losses, epoch, train_duration)
if tb_logger is not None:
tb_logger.save_losses(training_losses, epoch)
tb_logger.save_grads(epoch)
tb_logger.save_tensors(epoch)
first_iteration = True
validation_losses = defaultdict(AverageMeter)
logging.info('Running validation for epoch {}/{}'.format(
epoch, train_options.number_of_epochs))
#val
for image, _ in val_data:
image = image.to(device)
#crop imgs
imgs = cropImg(32, image)
#iterate img
bitwise_arr = []
main_losses = None
encoded_imgs = []
blocking_imgs = []
for img in imgs:
img = img.to(device)
message = torch.Tensor(
np.random.choice(
[0, 1], (img.shape[0],
hidden_config.message_length))).to(device)
losses, (encoded_images, noised_images,
decoded_messages) = model.validate_on_batch(
[img, message])
encoded_imgs.append(encoded_images)
blocking_imgs.append(
encoded_images[0][0].cpu().detach().numpy())
main_losses = losses
for name, loss in losses.items():
if (name == 'bitwise-error '):
bitwise_arr.append(loss)
Total = 0
Vcount = 0
V_average = 0
H_average = 0
for i in range(0, len(blocking_imgs) - 1):
if ((i + 1) % 4 != 0):
img = blocking_imgs[i]
img_next = blocking_imgs[i + 1]
average_img = 0
average_img_next = 0
for j in range(0, 32):
for k in range(0, 10):
average_img = average_img + img[j][31 - k]
average_img_next = average_img_next + img_next[j][k]
average_blocking = np.abs(average_img -
average_img_next) / 320
V_average = V_average + average_blocking
for j in range(0, 32):
distinct = np.abs(img[j][31] - img_next[j][0])
Total = Total + 1
if (distinct > 0.5):
Vcount = Vcount + 1
V_average = V_average / 12
Hcount = 0
for i in range(0, len(blocking_imgs) - 4):
img = blocking_imgs[i]
img_next = blocking_imgs[i + 4]
for j in range(0, 32):
for k in range(0, 10):
average_img = average_img + img[31 - k][j]
average_img_next = average_img_next + img_next[k][j]
average_blocking = np.abs(average_img - average_img_next) / 320
H_average = H_average + average_blocking
for j in range(0, 32):
distinct = np.abs(img[31][j] - img_next[0][j])
Total = Total + 1
if (distinct > 0.5):
Hcount = Hcount + 1
H_average = H_average / 12
bitwise_arr = np.array(bitwise_arr)
bitwise_avg = np.average(bitwise_arr)
#blocking_loss = (Vcount+Hcount)/Total
blocking_loss = (H_average + V_average) / 2
for name, loss in main_losses.items():
if (name == 'bitwise-error '):
validation_losses[name].update(bitwise_avg)
else:
if (name == 'blocking_effect'):
validation_losses[name].update(blocking_loss)
else:
validation_losses[name].update(loss)
#concat image
encoded_images = concatImgs(encoded_imgs)
if first_iteration:
if hidden_config.enable_fp16:
image = image.float()
encoded_images = encoded_images.float()
utils.save_images(
image.cpu()[:images_to_save, :, :, :],
encoded_images[:images_to_save, :, :, :].cpu(),
epoch,
os.path.join(this_run_folder, 'images'),
resize_to=saved_images_size)
first_iteration = False
utils.log_progress(validation_losses)
logging.info('-' * 40)
utils.save_checkpoint(model, train_options.experiment_name, epoch,
os.path.join(this_run_folder, 'checkpoints'))
utils.write_losses(os.path.join(this_run_folder, 'validation.csv'),
validation_losses, epoch,
time.time() - epoch_start)
def main():
device = torch.device(
'cuda') if torch.cuda.is_available() else torch.device('cpu')
parent_parser = argparse.ArgumentParser(
description='Training of HiDDeN nets')
subparsers = parent_parser.add_subparsers(dest='command',
help='Sub-parser for commands')
new_run_parser = subparsers.add_parser('new', help='starts a new run')
new_run_parser.add_argument('--data-dir',
'-d',
required=True,
type=str,
help='The directory where the data is stored.')
# Anno dir
new_run_parser.add_argument(
'--anno-dir',
'-a',
type=str,
help=
'The directory where the annotations are stored. Specify only if you have annotations in a different folder.'
)
new_run_parser.add_argument('--batch-size',
'-b',
required=True,
type=int,
help='The batch size.')
new_run_parser.add_argument('--epochs',
'-e',
default=300,
type=int,
help='Number of epochs to run the simulation.')
new_run_parser.add_argument('--name',
required=True,
type=str,
help='The name of the experiment.')
new_run_parser.add_argument(
'--size',
'-s',
default=128,
type=int,
help=
'The size of the images (images are square so this is height and width).'
)
new_run_parser.add_argument('--message',
'-m',
default=256,
type=int,
help='The length in bits of the watermark.')
new_run_parser.add_argument(
'--continue-from-folder',
'-c',
default='',
type=str,
help=
'The folder from where to continue a previous run. Leave blank if you are starting a new experiment.'
)
# parser.add_argument('--tensorboard', dest='tensorboard', action='store_true',
# help='If specified, use adds a Tensorboard log. On by default')
new_run_parser.add_argument('--tensorboard',
action='store_true',
help='Use to switch on Tensorboard logging.')
new_run_parser.add_argument('--enable-fp16',
dest='enable_fp16',
action='store_true',
help='Enable mixed-precision training.')
new_run_parser.add_argument(
'--noise',
nargs='*',
action=NoiseArgParser,
help=
"Noise layers configuration. Use quotes when specifying configuration, e.g. 'cropout((0.55, 0.6), (0.55, 0.6))'"
)
new_run_parser.set_defaults(tensorboard=False)
new_run_parser.set_defaults(enable_fp16=False)
new_run_parser.add_argument('--vocab-path',
'-v',
type=str,
default='./data/vocab.pkl',
help='load the vocab')
continue_parser = subparsers.add_parser('continue',
help='Continue a previous run')
continue_parser.add_argument(
'--folder',
'-f',
required=True,
type=str,
help='Continue from the last checkpoint in this folder.')
continue_parser.add_argument(
'--data-dir',
'-d',
required=False,
type=str,
help=
'The directory where the data is stored. Specify a value only if you want to override the previous value.'
)
# Anno dir
continue_parser.add_argument(
'--anno-dir',
'-a',
required=False,
type=str,
help=
'The directory where the annotations are stored. Specify a value only if you want to override the previous value.'
)
continue_parser.add_argument(
'--epochs',
'-e',
required=False,
type=int,
help=
'Number of epochs to run the simulation. Specify a value only if you want to override the previous value.'
)
args = parent_parser.parse_args()
checkpoint = None
loaded_checkpoint_file_name = None
with open(args.vocab_path, 'rb') as f:
vocab = pickle.load(f)
if args.command == 'continue':
this_run_folder = args.folder
options_file = os.path.join(this_run_folder,
'options-and-config.pickle')
train_options, hidden_config, noise_config = utils.load_options(
options_file)
checkpoint, loaded_checkpoint_file_name = utils.load_last_checkpoint(
os.path.join(this_run_folder, 'checkpoints'))
train_options.start_epoch = checkpoint['epoch'] + 1
if args.data_dir is not None:
train_options.train_folder = os.path.join(args.data_dir, 'train')
train_options.validation_folder = os.path.join(
args.data_dir, 'val')
if args.epochs is not None:
if train_options.start_epoch < args.epochs:
train_options.number_of_epochs = args.epochs
else:
print(
f'Command-line specifies of number of epochs = {args.epochs}, but folder={args.folder} '
f'already contains checkpoint for epoch = {train_options.start_epoch}.'
)
exit(1)
else:
assert args.command == 'new'
start_epoch = 1
train_options = TrainingOptions(
batch_size=args.batch_size,
number_of_epochs=args.epochs,
train_folder=os.path.join(args.data_dir, 'train'),
validation_folder=os.path.join(args.data_dir, 'val'),
ann_train=os.path.join(args.data_dir, 'ann_train.json'),
ann_val=os.path.join(args.data_dir, 'ann_val.json'),
runs_folder=os.path.join('.', 'runs'),
start_epoch=start_epoch,
experiment_name=args.name)
noise_config = args.noise if args.noise is not None else []
hidden_config = HiDDenConfiguration(H=args.size,
W=args.size,
message_length=args.message,
encoder_blocks=4,
encoder_channels=64,
decoder_blocks=7,
decoder_channels=64,
use_discriminator=True,
use_vgg=False,
discriminator_blocks=3,
discriminator_channels=64,
decoder_loss=1,
encoder_loss=0.7,
adversarial_loss=1e-3,
vocab_size=len(vocab),
enable_fp16=args.enable_fp16)
this_run_folder = utils.create_folder_for_run(
train_options.runs_folder, args.name)
with open(os.path.join(this_run_folder, 'options-and-config.pickle'),
'wb+') as f:
pickle.dump(train_options, f)
pickle.dump(noise_config, f)
pickle.dump(hidden_config, f)
logging.basicConfig(level=logging.INFO,
format='%(message)s',
handlers=[
logging.FileHandler(
os.path.join(
this_run_folder,
f'{train_options.experiment_name}.log')),
logging.StreamHandler(sys.stdout)
])
if (args.command == 'new' and args.tensorboard) or \
(args.command == 'continue' and os.path.isdir(os.path.join(this_run_folder, 'tb-logs'))):
logging.info('Tensorboard is enabled. Creating logger.')
from tensorboard_logger import TensorBoardLogger
tb_logger = TensorBoardLogger(os.path.join(this_run_folder, 'tb-logs'))
else:
tb_logger = None
noiser = Noiser(noise_config, device)
model = Hidden(hidden_config, device, noiser, tb_logger)
if args.command == 'continue':
# if we are continuing, we have to load the model params
assert checkpoint is not None
logging.info(
f'Loading checkpoint from file {loaded_checkpoint_file_name}')
utils.model_from_checkpoint(model, checkpoint)
logging.info('HiDDeN model: {}\n'.format(model.to_stirng()))
logging.info('Model Configuration:\n')
logging.info(pprint.pformat(vars(hidden_config)))
logging.info('\nNoise configuration:\n')
logging.info(pprint.pformat(str(noise_config)))
logging.info('\nTraining train_options:\n')
logging.info(pprint.pformat(vars(train_options)))
train(model, device, hidden_config, train_options, this_run_folder,
tb_logger, vocab)
def train(model: Hidden,
device: torch.device,
hidden_config: HiDDenConfiguration,
train_options: TrainingOptions,
this_run_folder: str,
tb_logger):
"""
Trains the HiDDeN model
:param model: The model
:param device: torch.device object, usually this is GPU (if avaliable), otherwise CPU.
:param hidden_config: The network configuration
:param train_options: The training settings
:param this_run_folder: The parent folder for the current training run to store training artifacts/results/logs.
:param tb_logger: TensorBoardLogger object which is a thin wrapper for TensorboardX logger.
Pass None to disable TensorboardX logging
:return:
"""
train_data, val_data = utils.get_data_loaders(hidden_config, train_options)
block_size = hidden_config.block_size
block_number = int(hidden_config.H/hidden_config.block_size)
val_folder = train_options.validation_folder
loss_type = train_options.loss_mode
m_length = hidden_config.message_length
alpha = train_options.alpha
img_names = listdir(val_folder+"/valid_class")
img_names.sort()
out_folder = train_options.output_folder
default = train_options.default
beta = train_options.beta
crop_width = int(beta*block_size)
file_count = len(train_data.dataset)
if file_count % train_options.batch_size == 0:
steps_in_epoch = file_count // train_options.batch_size
else:
steps_in_epoch = file_count // train_options.batch_size + 1
print_each = 10
images_to_save = 8
saved_images_size = (512, 512)
icount = 0
plot_block = []
for epoch in range(train_options.start_epoch, train_options.number_of_epochs + 1):
logging.info('\nStarting epoch {}/{}'.format(epoch, train_options.number_of_epochs))
logging.info('Batch size = {}\nSteps in epoch = {}'.format(train_options.batch_size, steps_in_epoch))
training_losses = defaultdict(AverageMeter)
epoch_start = time.time()
step = 1
#train
for image, _ in train_data:
image = image.to(device)
#crop imgs into blocks
imgs, modified_imgs, entropies = cropImg(block_size,image,crop_width,alpha)
bitwise_arr=[]
main_losses = None
encoded_imgs = []
batch = 0
for img, modified_img, entropy in zip(imgs,modified_imgs, entropies):
img=img.to(device)
modified_img = modified_img.to(device)
entropy = entropy.to(device)
message = torch.Tensor(np.random.choice([0, 1], (img.shape[0], m_length))).to(device)
losses, (encoded_images, noised_images, decoded_messages) = \
model.train_on_batch([img, message, modified_img, entropy,loss_type])
encoded_imgs.append(encoded_images)
batch = encoded_images.shape[0]
#get loss in the last block
if main_losses is None:
main_losses = losses
for k in losses:
main_losses[k] = losses[k]/len(imgs)
else:
for k in main_losses:
main_losses[k] += losses[k]/len(imgs)
#blocking effect loss calculation
blocking_loss = blocking_value(encoded_imgs,batch,block_size,block_number)
#update bitwise training loss
for name, loss in main_losses.items():
if(default == False and name == 'blocking_effect'):
training_losses[name].update(blocking_loss)
else:
training_losses[name].update(loss)
#statistic
if step % print_each == 0 or step == steps_in_epoch:
logging.info(
'Epoch: {}/{} Step: {}/{}'.format(epoch, train_options.number_of_epochs, step, steps_in_epoch))
utils.log_progress(training_losses)
logging.info('-' * 40)
step += 1
train_duration = time.time() - epoch_start
logging.info('Epoch {} training duration {:.2f} sec'.format(epoch, train_duration))
logging.info('-' * 40)
utils.write_losses(os.path.join(this_run_folder, 'train.csv'), training_losses, epoch, train_duration)
if tb_logger is not None:
tb_logger.save_losses(training_losses, epoch)
tb_logger.save_grads(epoch)
tb_logger.save_tensors(epoch)
first_iteration = True
validation_losses = defaultdict(AverageMeter)
logging.info('Running validation for epoch {}/{}'.format(epoch, train_options.number_of_epochs))
#validation
ep_blocking = 0
ep_total = 0
for image, _ in val_data:
image = image.to(device)
#crop imgs
imgs, modified_imgs, entropies = cropImg(block_size,image,crop_width,alpha)
bitwise_arr=[]
main_losses = None
encoded_imgs = []
batch = 0
for img, modified_img, entropy in zip(imgs,modified_imgs, entropies):
img=img.to(device)
modified_img = modified_img.to(device)
entropy = entropy.to(device)
message = torch.Tensor(np.random.choice([0, 1], (img.shape[0], m_length))).to(device)
losses, (encoded_images, noised_images, decoded_messages) = \
model.train_on_batch([img, message, modified_img, entropy,loss_type])
encoded_imgs.append(encoded_images)
batch = encoded_images.shape[0]
#get loss in the last block
if main_losses is None:
main_losses = losses
for k in losses:
main_losses[k] = losses[k]/len(imgs)
else:
for k in main_losses:
main_losses[k] += losses[k]/len(imgs)
#blocking value for plotting
blocking_loss = blocking_value(encoded_imgs,batch,block_size,block_number)
ep_blocking = ep_blocking+ blocking_loss
ep_total = ep_total+1
for name, loss in main_losses.items():
if(default == False and name == 'blocking_effect'):
validation_losses[name].update(blocking_loss)
else:
validation_losses[name].update(loss)
#concat image
encoded_images = concatImgs(encoded_imgs,block_number)
#save_image(encoded_images,"enc_img"+str(epoch)+".png")
#save_image(image,"original_img"+str(epoch)+".png")
if first_iteration:
if hidden_config.enable_fp16:
image = image.float()
encoded_images = encoded_images.float()
utils.save_images(image.cpu()[:images_to_save, :, :, :],
encoded_images[:images_to_save, :, :, :].cpu(),
epoch,
os.path.join(this_run_folder, 'images'), resize_to=saved_images_size)
first_iteration = False
#save validation in the last epoch
if(epoch == train_options.number_of_epochs):
if(train_options.ats):
for i in range(0,batch):
image = encoded_images[i].cpu()
image = (image + 1) / 2
f_dst = out_folder+"/"+img_names[icount]
save_image(image,f_dst)
icount = icount+1
#append block effect for plotting
plot_block.append(ep_blocking/ep_total)
utils.log_progress(validation_losses)
logging.info('-' * 40)
utils.save_checkpoint(model, train_options.experiment_name, epoch, os.path.join(this_run_folder, 'checkpoints'))
utils.write_losses(os.path.join(this_run_folder, 'validation.csv'), validation_losses, epoch,
time.time() - epoch_start)
def train_own_noise(model: Hidden, device: torch.device,
hidden_config: HiDDenConfiguration,
train_options: TrainingOptions, this_run_folder: str,
tb_logger, noise):
"""
Trains the HiDDeN model
:param model: The model
:param device: torch.device object, usually this is GPU (if avaliable), otherwise CPU.
:param hidden_config: The network configuration
:param train_options: The training settings
:param this_run_folder: The parent folder for the current training run to store training artifacts/results/logs.
:param tb_logger: TensorBoardLogger object which is a thin wrapper for TensorboardX logger.
Pass None to disable TensorboardX logging
:return:
"""
train_data, val_data = utils.get_data_loaders(hidden_config, train_options)
file_count = len(train_data.dataset)
if file_count % train_options.batch_size == 0:
steps_in_epoch = file_count // train_options.batch_size
else:
steps_in_epoch = file_count // train_options.batch_size + 1
steps_in_epoch = 313
print_each = 10
images_to_save = 8
saved_images_size = (
512, 512) # for qualitative check purpose to use a larger size
for epoch in range(train_options.start_epoch,
train_options.number_of_epochs + 1):
logging.info('\nStarting epoch {}/{}'.format(
epoch, train_options.number_of_epochs))
logging.info('Batch size = {}\nSteps in epoch = {}'.format(
train_options.batch_size, steps_in_epoch))
training_losses = defaultdict(AverageMeter)
if train_options.video_dataset:
random.shuffle(train_data.dataset)
epoch_start = time.time()
step = 1
for image, _ in train_data:
image = image.to(device)
message = torch.Tensor(
np.random.choice(
[0, 1],
(image.shape[0], hidden_config.message_length))).to(device)
losses, _ = model.train_on_batch([image, message])
for name, loss in losses.items():
training_losses[name].update(loss)
if step % print_each == 0 or step == steps_in_epoch:
#import pdb; pdb.set_trace()
logging.info('Epoch: {}/{} Step: {}/{}'.format(
epoch, train_options.number_of_epochs, step,
steps_in_epoch))
utils.log_progress(training_losses)
logging.info('-' * 40)
step += 1
if step == steps_in_epoch:
break
train_duration = time.time() - epoch_start
logging.info('Epoch {} training duration {:.2f} sec'.format(
epoch, train_duration))
logging.info('-' * 40)
utils.write_losses(os.path.join(this_run_folder, 'train.csv'),
training_losses, epoch, train_duration)
if tb_logger is not None:
tb_logger.save_losses(training_losses, epoch)
tb_logger.save_grads(epoch)
tb_logger.save_tensors(epoch)
first_iteration = True
validation_losses = defaultdict(AverageMeter)
logging.info('Running validation for epoch {}/{} for noise {}'.format(
epoch, train_options.number_of_epochs, noise))
step = 1
for image, _ in val_data:
image = image.to(device)
message = torch.Tensor(
np.random.choice(
[0, 1],
(image.shape[0], hidden_config.message_length))).to(device)
losses, (
encoded_images, noised_images,
decoded_messages) = model.validate_on_batch_specific_noise(
[image, message], noise=noise)
for name, loss in losses.items():
validation_losses[name].update(loss)
if first_iteration:
if hidden_config.enable_fp16:
image = image.float()
encoded_images = encoded_images.float()
utils.save_images(
image.cpu()[:images_to_save, :, :, :],
encoded_images[:images_to_save, :, :, :].cpu(),
epoch,
os.path.join(this_run_folder, 'images'),
resize_to=saved_images_size)
first_iteration = False
step += 1
if step == steps_in_epoch // 10:
break
utils.log_progress(validation_losses)
logging.info('-' * 40)
utils.save_checkpoint(model, train_options.experiment_name, epoch,
os.path.join(this_run_folder, 'checkpoints'))
utils.write_losses(
os.path.join(this_run_folder, 'validation_' + noise + '.csv'),
validation_losses, epoch,
time.time() - epoch_start)
def train(model: Hidden, device: torch.device,
hidden_config: HiDDenConfiguration, train_options: TrainingOptions,
this_run_folder: str, tb_logger, vocab):
"""
Trains the HiDDeN model
:param model: The model
:param device: torch.device object, usually this is GPU (if avaliable), otherwise CPU.
:param hidden_config: The network configuration
:param train_options: The training settings
:param this_run_folder: The parent folder for the current training run to store training artifacts/results/logs.
:param tb_logger: TensorBoardLogger object which is a thin wrapper for TensorboardX logger.
Pass None to disable TensorboardX logging
:return:
"""
train_data, val_data = utils.get_data_loaders(hidden_config, train_options,
vocab)
file_count = len(train_data.dataset)
if file_count % train_options.batch_size == 0:
steps_in_epoch = file_count // train_options.batch_size
else:
steps_in_epoch = file_count // train_options.batch_size + 1
print_each = 10
images_to_save = 8
saved_images_size = (512, 512)
for epoch in range(train_options.start_epoch,
train_options.number_of_epochs + 1):
logging.info('\nStarting epoch {}/{}'.format(
epoch, train_options.number_of_epochs))
logging.info('Batch size = {}\nSteps in epoch = {}'.format(
train_options.batch_size, steps_in_epoch))
training_losses = defaultdict(AverageMeter)
epoch_start = time.time()
step = 1
for image, ekeys, dkeys, caption, length in train_data:
image, caption, ekeys, dkeys = image.to(device), caption.to(
device), ekeys.to(device), dkeys.to(device)
losses, _ = model.train_on_batch(
[image, ekeys, dkeys, caption, length])
for name, loss in losses.items():
training_losses[name].update(loss)
if step % print_each == 0 or step == steps_in_epoch:
logging.info('Epoch: {}/{} Step: {}/{}'.format(
epoch, train_options.number_of_epochs, step,
steps_in_epoch))
utils.log_progress(training_losses)
logging.info('-' * 40)
step += 1
train_duration = time.time() - epoch_start
logging.info('Epoch {} training duration {:.2f} sec'.format(
epoch, train_duration))
logging.info('-' * 40)
utils.write_losses(os.path.join(this_run_folder, 'train.csv'),
training_losses, epoch, train_duration)
if tb_logger is not None:
tb_logger.save_losses(training_losses, epoch)
tb_logger.save_grads(epoch)
tb_logger.save_tensors(epoch)
first_iteration = True
validation_losses = defaultdict(AverageMeter)
logging.info('Running validation for epoch {}/{}'.format(
epoch, train_options.number_of_epochs))
for image, ekeys, dkeys, caption, length in val_data:
image, caption, ekeys, dkeys = image.to(device), caption.to(
device), ekeys.to(device), dkeys.to(device)
losses, (encoded_images, noised_images, decoded_messages, predicted_sents) = \
model.validate_on_batch([image, ekeys, dkeys, caption, length])
#print(predicted)
#exit()
predicted_sents = predicted_sents.cpu().numpy()
for i in range(train_options.batch_size):
try:
#print(''.join([vocab.idx2word[int(w)] + ' ' for w in predicted.cpu().numpy()[i::train_options.batch_size]][1:length[i]-1]))
print("".join([
vocab.idx2word[int(idx)] + ' '
for idx in predicted_sents[i]
]))
break
except IndexError:
print(f'{i}th batch does not have enough length.')
for name, loss in losses.items():
validation_losses[name].update(loss)
if first_iteration:
if hidden_config.enable_fp16:
image = image.float()
encoded_images = encoded_images.float()
utils.save_images(
image.cpu()[:images_to_save, :, :, :],
encoded_images[:images_to_save, :, :, :].cpu(),
epoch,
os.path.join(this_run_folder, 'images'),
resize_to=saved_images_size)
first_iteration = False
utils.log_progress(validation_losses)
logging.info('-' * 40)
utils.save_checkpoint(model, train_options.experiment_name, epoch,
os.path.join(this_run_folder, 'checkpoints'))
utils.write_losses(os.path.join(this_run_folder, 'validation.csv'),
validation_losses, epoch,
time.time() - epoch_start)
def train(model: Hidden, device: torch.device,
hidden_config: HiDDenConfiguration, train_options: TrainingOptions,
this_run_folder: str, tb_logger):
"""
Trains the HiDDeN model
:param model: The model
:param device: torch.device object, usually this is GPU (if avaliable), otherwise CPU.
:param hidden_config: The network configuration
:param train_options: The training settings
:param this_run_folder: The parent folder for the current training run to store training artifacts/results/logs.
:param tb_logger: TensorBoardLogger object which is a thin wrapper for TensorboardX logger.
Pass None to disable TensorboardX logging
:return:
"""
train_data, val_data = utils.get_data_loaders(hidden_config, train_options)
file_count = len(train_data.dataset)
if file_count % train_options.batch_size == 0:
steps_in_epoch = file_count // train_options.batch_size
else:
steps_in_epoch = file_count // train_options.batch_size + 1
print_each = 10
images_to_save = 8
saved_images_size = (512, 512)
for epoch in range(train_options.start_epoch,
train_options.number_of_epochs + 1):
print('\nStarting epoch {}/{}'.format(epoch,
train_options.number_of_epochs))
print('Batch size = {}\nSteps in epoch = {}'.format(
train_options.batch_size, steps_in_epoch))
losses_accu = {}
epoch_start = time.time()
step = 1
for image, _ in train_data:
image = image.to(device)
message = torch.Tensor(
np.random.choice(
[0, 1],
(image.shape[0], hidden_config.message_length))).to(device)
losses, _ = model.train_on_batch([image, message])
if not losses_accu: # dict is empty, initialize
for name in losses:
losses_accu[name] = []
for name, loss in losses.items():
losses_accu[name].append(loss)
if step % print_each == 0 or step == steps_in_epoch:
print('Epoch: {}/{} Step: {}/{}'.format(
epoch, train_options.number_of_epochs, step,
steps_in_epoch))
utils.print_progress(losses_accu)
print('-' * 40)
step += 1
train_duration = time.time() - epoch_start
print('Epoch {} training duration {:.2f} sec'.format(
epoch, train_duration))
print('-' * 40)
utils.write_losses(os.path.join(this_run_folder, 'train.csv'),
losses_accu, epoch, train_duration)
if tb_logger is not None:
tb_logger.save_losses(losses_accu, epoch)
tb_logger.save_grads(epoch)
tb_logger.save_tensors(epoch)
first_iteration = True
print('Running validation for epoch {}/{}'.format(
epoch, train_options.number_of_epochs))
for image, _ in val_data:
image = image.to(device)
message = torch.Tensor(
np.random.choice(
[0, 1],
(image.shape[0], hidden_config.message_length))).to(device)
losses, (encoded_images, noised_images,
decoded_messages) = model.validate_on_batch(
[image, message])
if not losses_accu: # dict is empty, initialize
for name in losses:
losses_accu[name] = []
for name, loss in losses.items():
losses_accu[name].append(loss)
if first_iteration:
utils.save_images(
image.cpu()[:images_to_save, :, :, :],
encoded_images[:images_to_save, :, :, :].cpu(),
epoch,
os.path.join(this_run_folder, 'images'),
resize_to=saved_images_size)
first_iteration = False
utils.print_progress(losses_accu)
print('-' * 40)
utils.save_checkpoint(model, epoch, losses_accu,
os.path.join(this_run_folder, 'checkpoints'))
utils.write_losses(os.path.join(this_run_folder, 'validation.csv'),
losses_accu, epoch,
time.time() - epoch_start)
def main():
device = torch.device(
'cuda') if torch.cuda.is_available() else torch.device('cpu')
parser = argparse.ArgumentParser(description='Training of HiDDeN nets')
parser.add_argument('--size', '-s', default=128, type=int)
parser.add_argument('--data-dir', '-d', required=True, type=str)
parser.add_argument('--runs-folder',
'-sf',
default=os.path.join('.', 'runs'),
type=str)
parser.add_argument('--message', '-m', default=30, type=int)
parser.add_argument('--epochs', '-e', default=400, type=int)
parser.add_argument('--batch-size', '-b', required=True, type=int)
parser.add_argument('--continue-from-folder', '-c', default='', type=str)
parser.add_argument('--tensorboard',
dest='tensorboard',
action='store_true')
parser.add_argument('--no-tensorboard',
dest='tensorboard',
action='store_false')
parser.set_defaults(tensorboard=True)
args = parser.parse_args()
checkpoint = None
if args.continue_from_folder != '':
this_run_folder = args.continue_from_folder
train_options, hidden_config, noise_config = utils.load_options(
this_run_folder)
checkpoint = utils.load_last_checkpoint(
os.path.join(this_run_folder, 'checkpoints'))
train_options.start_epoch = checkpoint['epoch']
else:
start_epoch = 1
train_options = TrainingOptions(
batch_size=args.batch_size,
number_of_epochs=args.epochs,
train_folder=os.path.join(args.data_dir, 'train'),
validation_folder=os.path.join(args.data_dir, 'val'),
runs_folder=os.path.join('.', 'runs'),
start_epoch=start_epoch)
# noise_config = [
# {
# 'type': 'resize',
# 'resize_ratio': 0.4
# }]
noise_config = []
hidden_config = HiDDenConfiguration(H=args.size,
W=args.size,
message_length=args.message,
encoder_blocks=4,
encoder_channels=64,
decoder_blocks=7,
decoder_channels=64,
use_discriminator=True,
use_vgg=False,
discriminator_blocks=3,
discriminator_channels=64,
decoder_loss=1,
encoder_loss=0.7,
adversarial_loss=1e-3)
this_run_folder = utils.create_folder_for_run(train_options)
with open(os.path.join(this_run_folder, 'options-and-config.pickle'),
'wb+') as f:
pickle.dump(train_options, f)
pickle.dump(noise_config, f)
pickle.dump(hidden_config, f)
noiser = Noiser(noise_config, device)
if args.tensorboard:
print('Tensorboard is enabled. Creating logger.')
from tensorboard_logger import TensorBoardLogger
tb_logger = TensorBoardLogger(os.path.join(this_run_folder, 'tb-logs'))
else:
tb_logger = None
model = Hidden(hidden_config, device, noiser, tb_logger)
if args.continue_from_folder != '':
# if we are continuing, we have to load the model params
assert checkpoint is not None
utils.model_from_checkpoint(model, checkpoint)
print('HiDDeN model: {}\n'.format(model.to_stirng()))
print('Model Configuration:\n')
pprint.pprint(vars(hidden_config))
print('\nNoise configuration:\n')
pprint.pprint(str(noise_config))
print('\nTraining train_options:\n')
pprint.pprint(vars(train_options))
print()
train(model, device, hidden_config, train_options, this_run_folder,
tb_logger)
def main():
device = torch.device(
'cuda') if torch.cuda.is_available() else torch.device('cpu')
parent_parser = argparse.ArgumentParser(
description='Training of HiDDeN nets')
subparsers = parent_parser.add_subparsers(dest='command',
help='Sub-parser for commands')
new_run_parser = subparsers.add_parser('new', help='starts a new run')
new_run_parser.add_argument('--data-dir',
'-d',
required=True,
type=str,
help='The directory where the data is stored.')
new_run_parser.add_argument('--batch-size',
'-b',
required=True,
type=int,
help='The batch size.')
new_run_parser.add_argument('--epochs',
'-e',
default=300,
type=int,
help='Number of epochs to run the simulation.')
new_run_parser.add_argument('--name',
required=True,
type=str,
help='The name of the experiment.')
new_run_parser.add_argument('--adv_loss',
default=0,
required=False,
type=float,
help='Coefficient of the adversarial loss.')
new_run_parser.add_argument('--residual',
default=0,
required=False,
type=int,
help='If to use residual or not.')
new_run_parser.add_argument('--video_dataset',
default=0,
required=False,
type=int,
help='If to use video dataset or not.')
new_run_parser.add_argument(
'--save-dir',
'-sd',
default='runs',
required=True,
type=str,
help='The save directory where the result is stored.')
new_run_parser.add_argument(
'--size',
'-s',
default=128,
type=int,
help=
'The size of the images (images are square so this is height and width).'
)
new_run_parser.add_argument('--message',
'-m',
default=30,
type=int,
help='The length in bits of the watermark.')
new_run_parser.add_argument(
'--continue-from-folder',
'-c',
default='',
type=str,
help=
'The folder from where to continue a previous run. Leave blank if you are starting a new experiment.'
)
# parser.add_argument('--tensorboard', dest='tensorboard', action='store_true',
# help='If specified, use adds a Tensorboard log. On by default')
new_run_parser.add_argument('--tensorboard',
action='store_true',
help='Use to switch on Tensorboard logging.')
new_run_parser.add_argument('--enable-fp16',
dest='enable_fp16',
action='store_true',
help='Enable mixed-precision training.')
new_run_parser.add_argument(
'--noise',
nargs='*',
action=NoiseArgParser,
help=
"Noise layers configuration. Use quotes when specifying configuration, e.g. 'cropout((0.55, 0.6), (0.55, 0.6))'"
)
new_run_parser.add_argument('--hostname',
default=socket.gethostname(),
help='the host name of the running server')
new_run_parser.add_argument(
'--cover-dependent',
default=1,
required=False,
type=int,
help='If to use cover dependent architecture or not.')
new_run_parser.add_argument('--jpeg_type',
'-j',
required=False,
type=str,
default='jpeg',
help='Jpeg type used in the combined2 noise.')
new_run_parser.set_defaults(tensorboard=False)
new_run_parser.set_defaults(enable_fp16=False)
continue_parser = subparsers.add_parser('continue',
help='Continue a previous run')
continue_parser.add_argument(
'--folder',
'-f',
required=True,
type=str,
help='Continue from the last checkpoint in this folder.')
continue_parser.add_argument(
'--data-dir',
'-d',
required=False,
type=str,
help=
'The directory where the data is stored. Specify a value only if you want to override the previous value.'
)
continue_parser.add_argument(
'--epochs',
'-e',
required=False,
type=int,
help=
'Number of epochs to run the simulation. Specify a value only if you want to override the previous value.'
)
# continue_parser.add_argument('--tensorboard', action='store_true',
# help='Override the previous setting regarding tensorboard logging.')
# Setting up a seed for debug
seed = 123
torch.manual_seed(seed)
np.random.seed(seed)
args = parent_parser.parse_args()
checkpoint = None
loaded_checkpoint_file_name = None
print(args.cover_dependent)
if not args.video_dataset:
if args.hostname == 'ee898-System-Product-Name':
args.data_dir = '/home/ee898/Desktop/chaoning/ImageNet'
args.hostname = 'ee898'
elif args.hostname == 'DL178':
args.data_dir = '/media/user/SSD1TB-2/ImageNet'
else:
args.data_dir = '/workspace/data_local/imagenet_pytorch'
else:
if args.hostname == 'ee898-System-Product-Name':
args.data_dir = '/home/ee898/Desktop/chaoning/ImageNet'
args.hostname = 'ee898'
elif args.hostname == 'DL178':
args.data_dir = '/media/user/SSD1TB-2/ImageNet'
else:
args.data_dir = './oops_dataset/oops_video'
assert args.data_dir
if args.command == 'continue':
this_run_folder = args.folder
options_file = os.path.join(this_run_folder,
'options-and-config.pickle')
train_options, hidden_config, noise_config = utils.load_options(
options_file)
checkpoint, loaded_checkpoint_file_name = utils.load_last_checkpoint(
os.path.join(this_run_folder, 'checkpoints'))
train_options.start_epoch = checkpoint['epoch'] + 1
if args.data_dir is not None:
train_options.train_folder = os.path.join(args.data_dir, 'train')
train_options.validation_folder = os.path.join(
args.data_dir, 'val')
if args.epochs is not None:
if train_options.start_epoch < args.epochs:
train_options.number_of_epochs = args.epochs
else:
print(
f'Command-line specifies of number of epochs = {args.epochs}, but folder={args.folder} '
f'already contains checkpoint for epoch = {train_options.start_epoch}.'
)
exit(1)
else:
assert args.command == 'new'
start_epoch = 1
train_options = TrainingOptions(
batch_size=args.batch_size,
number_of_epochs=args.epochs,
train_folder=os.path.join(args.data_dir, 'train'),
validation_folder=os.path.join(args.data_dir, 'val'),
runs_folder=os.path.join('.', args.save_dir),
start_epoch=start_epoch,
experiment_name=args.name,
video_dataset=args.video_dataset)
noise_config = args.noise if args.noise is not None else []
hidden_config = HiDDenConfiguration(
H=args.size,
W=args.size,
message_length=args.message,
encoder_blocks=4,
encoder_channels=64,
decoder_blocks=7,
decoder_channels=64,
use_discriminator=True,
use_vgg=False,
discriminator_blocks=3,
discriminator_channels=64,
decoder_loss=1,
encoder_loss=0.7,
adversarial_loss=args.adv_loss,
cover_dependent=args.cover_dependent,
residual=args.residual,
enable_fp16=args.enable_fp16)
this_run_folder = utils.create_folder_for_run(
train_options.runs_folder, args.name)
with open(os.path.join(this_run_folder, 'options-and-config.pickle'),
'wb+') as f:
pickle.dump(train_options, f)
pickle.dump(noise_config, f)
pickle.dump(hidden_config, f)
logging.basicConfig(level=logging.INFO,
format='%(message)s',
handlers=[
logging.FileHandler(
os.path.join(
this_run_folder,
f'{train_options.experiment_name}.log')),
logging.StreamHandler(sys.stdout)
])
if (args.command == 'new' and args.tensorboard) or \
(args.command == 'continue' and os.path.isdir(os.path.join(this_run_folder, 'tb-logs'))):
logging.info('Tensorboard is enabled. Creating logger.')
from tensorboard_logger import TensorBoardLogger
tb_logger = TensorBoardLogger(os.path.join(this_run_folder, 'tb-logs'))
else:
tb_logger = None
noiser = Noiser(noise_config, device, args.jpeg_type)
model = Hidden(hidden_config, device, noiser, tb_logger)
if args.command == 'continue':
# if we are continuing, we have to load the model params
assert checkpoint is not None
logging.info(
f'Loading checkpoint from file {loaded_checkpoint_file_name}')
utils.model_from_checkpoint(model, checkpoint)
logging.info('HiDDeN model: {}\n'.format(model.to_stirng()))
logging.info('Model Configuration:\n')
logging.info(pprint.pformat(vars(hidden_config)))
logging.info('\nNoise configuration:\n')
logging.info(pprint.pformat(str(noise_config)))
logging.info('\nTraining train_options:\n')
logging.info(pprint.pformat(vars(train_options)))
# train(model, device, hidden_config, train_options, this_run_folder, tb_logger)
# train_other_noises(model, device, hidden_config, train_options, this_run_folder, tb_logger)
if str(args.noise[0]) == "WebP()":
noise = 'webp'
elif str(args.noise[0]) == "JpegCompression2000()":
noise = 'jpeg2000'
elif str(args.noise[0]) == "MPEG4()":
noise = 'mpeg4'
elif str(args.noise[0]) == "H264()":
noise = 'h264'
elif str(args.noise[0]) == "XVID()":
noise = 'xvid'
elif str(args.noise[0]) == "DiffQFJpegCompression2()":
noise = 'diff_qf_jpeg2'
elif str(args.noise[0]) == "DiffCorruptions()":
noise = 'diff_corruptions'
else:
noise = 'jpeg'
train_own_noise(model, device, hidden_config, train_options,
this_run_folder, tb_logger, noise)
def main():
# device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
device = torch.device('cpu')
parser = argparse.ArgumentParser(description='Training of HiDDeN nets')
# parser.add_argument('--size', '-s', default=128, type=int, help='The size of the images (images are square so this is height and width).')
parser.add_argument('--data-dir',
'-d',
required=True,
type=str,
help='The directory where the data is stored.')
parser.add_argument(
'--runs_root',
'-r',
default=os.path.join('.', 'experiments'),
type=str,
help='The root folder where data about experiments are stored.')
args = parser.parse_args()
print_each = 25
completed_runs = [
o for o in os.listdir(args.runs_root)
if os.path.isdir(os.path.join(args.runs_root, o))
and o != 'no-noise-defaults'
]
print(completed_runs)
write_csv_header = True
for run_name in completed_runs:
current_run = os.path.join(args.runs_root, run_name)
print(f'Run folder: {current_run}')
options_file = os.path.join(current_run, 'options-and-config.pickle')
train_options, hidden_config, noise_config = utils.load_options(
options_file)
train_options.train_folder = os.path.join(args.data_dir, 'val')
train_options.validation_folder = os.path.join(args.data_dir, 'val')
train_options.batch_size = 4
checkpoint = utils.load_last_checkpoint(
os.path.join(current_run, 'checkpoints'))
noiser = Noiser(noise_config, device)
model = Hidden(hidden_config, device, noiser, tb_logger=None)
utils.model_from_checkpoint(model, checkpoint)
print('Model loaded successfully. Starting validation run...')
_, val_data = utils.get_data_loaders(hidden_config, train_options)
file_count = len(val_data.dataset)
if file_count % train_options.batch_size == 0:
steps_in_epoch = file_count // train_options.batch_size
else:
steps_in_epoch = file_count // train_options.batch_size + 1
losses_accu = {}
step = 0
for image, _ in val_data:
step += 1
image = image.to(device)
message = torch.Tensor(
np.random.choice(
[0, 1],
(image.shape[0], hidden_config.message_length))).to(device)
losses, (encoded_images, noised_images,
decoded_messages) = model.validate_on_batch(
[image, message])
if not losses_accu: # dict is empty, initialize
for name in losses:
losses_accu[name] = []
for name, loss in losses.items():
losses_accu[name].append(loss)
if step % print_each == 0:
print(f'Step {step}/{steps_in_epoch}')
utils.print_progress(losses_accu)
print('-' * 40)
utils.print_progress(losses_accu)
write_validation_loss(os.path.join(args.runs_root,
'validation_run.csv'),
losses_accu,
run_name,
checkpoint['epoch'],
write_header=write_csv_header)
write_csv_header = False
def main():
device = torch.device(
'cuda') if torch.cuda.is_available() else torch.device('cpu')
parser = argparse.ArgumentParser(description='Training of HiDDeN nets')
parser.add_argument('--data-dir',
'-d',
required=True,
type=str,
help='The directory where the data is stored.')
parser.add_argument('--batch-size',
'-b',
required=True,
type=int,
help='The batch size.')
parser.add_argument('--epochs',
'-e',
default=400,
type=int,
help='Number of epochs to run the simulation.')
parser.add_argument('--name',
required=True,
type=str,
help='The name of the experiment.')
parser.add_argument(
'--runs-folder',
'-sf',
default=os.path.join('.', 'runs'),
type=str,
help='The root folder where data about experiments are stored.')
parser.add_argument(
'--size',
'-s',
default=128,
type=int,
help=
'The size of the images (images are square so this is height and width).'
)
parser.add_argument('--message',
'-m',
default=30,
type=int,
help='The length in bits of the watermark.')
parser.add_argument(
'--continue-from-folder',
'-c',
default='',
type=str,
help=
'The folder from where to continue a previous run. Leave blank if you are starting a new experiment.'
)
parser.add_argument(
'--tensorboard',
dest='tensorboard',
action='store_true',
help='If specified, use adds a Tensorboard log. On by default')
parser.add_argument('--no-tensorboard',
dest='tensorboard',
action='store_false',
help='Use to switch off Tensorboard logging.')
parser.add_argument(
'--noise',
nargs='*',
action=NoiseArgParser,
help=
"Noise layers configuration. Use quotes when specifying configuration, e.g. 'cropout((0.55, 0.6), (0.55, 0.6))'"
)
parser.set_defaults(tensorboard=True)
args = parser.parse_args()
checkpoint = None
if args.continue_from_folder != '':
this_run_folder = args.continue_from_folder
options_file = os.path.join(this_run_folder,
'options-and-config.pickle')
train_options, hidden_config, noise_config = utils.load_options(
options_file)
checkpoint = utils.load_last_checkpoint(
os.path.join(this_run_folder, 'checkpoints'))
train_options.start_epoch = checkpoint['epoch'] + 1
else:
start_epoch = 1
train_options = TrainingOptions(
batch_size=args.batch_size,
number_of_epochs=args.epochs,
train_folder=os.path.join(args.data_dir, 'train'),
validation_folder=os.path.join(args.data_dir, 'val'),
runs_folder=os.path.join('.', 'runs'),
start_epoch=start_epoch,
experiment_name=args.name)
noise_config = args.noise if args.noise is not None else []
hidden_config = HiDDenConfiguration(H=args.size,
W=args.size,
message_length=args.message,
encoder_blocks=4,
encoder_channels=64,
decoder_blocks=7,
decoder_channels=64,
use_discriminator=True,
use_vgg=False,
discriminator_blocks=3,
discriminator_channels=64,
decoder_loss=1,
encoder_loss=0.7,
adversarial_loss=1e-3)
this_run_folder = utils.create_folder_for_run(
train_options.runs_folder, args.name)
with open(os.path.join(this_run_folder, 'options-and-config.pickle'),
'wb+') as f:
pickle.dump(train_options, f)
pickle.dump(noise_config, f)
pickle.dump(hidden_config, f)
logging.basicConfig(level=logging.INFO,
format='%(message)s',
handlers=[
logging.FileHandler(
os.path.join(this_run_folder,
f'{args.name}.log')),
logging.StreamHandler(sys.stdout)
])
noiser = Noiser(noise_config, device)
if args.tensorboard:
logging.info('Tensorboard is enabled. Creating logger.')
from tensorboard_logger import TensorBoardLogger
tb_logger = TensorBoardLogger(os.path.join(this_run_folder, 'tb-logs'))
else:
tb_logger = None
model = Hidden(hidden_config, device, noiser, tb_logger)
if args.continue_from_folder != '':
# if we are continuing, we have to load the model params
assert checkpoint is not None
utils.model_from_checkpoint(model, checkpoint)
logging.info('HiDDeN model: {}\n'.format(model.to_stirng()))
logging.info('Model Configuration:\n')
logging.info(pprint.pformat(vars(hidden_config)))
logging.info('\nNoise configuration:\n')
logging.info(pprint.pformat(str(noise_config)))
logging.info('\nTraining train_options:\n')
logging.info(pprint.pformat(vars(train_options)))
train(model, device, hidden_config, train_options, this_run_folder,
tb_logger)
def main():
device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
parent_parser = argparse.ArgumentParser(description='Training of HiDDeN nets')
subparsers = parent_parser.add_subparsers(dest='command', help='Sub-parser for commands')
new_run_parser = subparsers.add_parser('new', help='starts a new run')
new_run_parser.add_argument('--data-dir', '-d', required=True, type=str,
help='The directory where the data is stored.')
new_run_parser.add_argument('--batch-size', '-b', default=30, type=int, help='The batch size.')
new_run_parser.add_argument('--epochs', '-e', default=300, type=int, help='Number of epochs to run the simulation.')
new_run_parser.add_argument('--name', required=True, type=str, help='The name of the experiment.')
new_run_parser.add_argument('--size', '-s', default=128, type=int, help='The size of the images (images are square so this is height and width).')
new_run_parser.add_argument('--in_channels', default=3, type=int, help='input channel size')
new_run_parser.add_argument('--message', '-m', default=32, type=int, help='The length in bits of the watermark.')
new_run_parser.add_argument('--ratio', default=0.2, type=float, help='ratio of dataset.')
new_run_parser.add_argument('--continue-from-folder', '-c', default='', type=str,
help='The folder from where to continue a previous run. Leave blank if you are starting a new experiment.')
new_run_parser.add_argument('--enable-fp16', dest='enable_fp16', action='store_true',
help='Enable mixed-precision training.')
new_run_parser.add_argument('--noise', nargs='*', action=NoiseArgParser,
help="Noise layers configuration. Use quotes when specifying configuration, e.g. 'cropout((0.55, 0.6), (0.55, 0.6))'")
new_run_parser.set_defaults(enable_fp16=False)
continue_parser = subparsers.add_parser('continue', help='Continue a previous run')
continue_parser.add_argument('--folder', '-f', required=True, type=str,
help='Continue from the last checkpoint in this folder.')
continue_parser.add_argument('--data-dir', '-d', required=False, type=str,
help='The directory where the data is stored. Specify a value only if you want to override the previous value.')
continue_parser.add_argument('--epochs', '-e', required=False, type=int,
help='Number of epochs to run the simulation. Specify a value only if you want to override the previous value.')
args = parent_parser.parse_args()
checkpoint = None
loaded_checkpoint_file_name = None
if args.command == 'continue':
options_file = os.path.join(args.folder, 'options-and-config.pickle')
train_options, hidden_config, noise_config = utils.load_options(options_file)
checkpoint, loaded_checkpoint_file_name = utils.load_last_checkpoint(os.path.join(args.folder, 'checkpoints'))
train_options.start_epoch = checkpoint['epoch'] + 1
train_options.best_epoch = checkpoint['best_epoch']
train_options.best_cond = checkpoint['best_cond']
if args.epochs is not None:
if train_options.start_epoch < args.epochs:
train_options.number_of_epochs = args.epochs
else:
print(f'Command-line specifies of number of epochs = {args.epochs}, but folder={args.folder} '
f'already contains checkpoint for epoch = {train_options.start_epoch}.')
exit(1)
else:
assert args.command == 'new'
start_epoch = 1
train_options = TrainingOptions(
batch_size=args.batch_size,
number_of_epochs=args.epochs, data_ratio=args.ratio,
data_dir=args.data_dir,
runs_folder='./runs', tb_logger_folder='./logger',
start_epoch=start_epoch, experiment_name=f'{args.name}_r{int(100*args.ratio):03d}b{args.size}ch{args.in_channels}m{args.message}')
noise_config = args.noise if args.noise is not None else []
hidden_config = HiDDenConfiguration(H=args.size, W=args.size,input_channels=args.in_channels,
message_length=args.message,
encoder_blocks=4, encoder_channels=64,
decoder_blocks=7, decoder_channels=64,
use_discriminator=True,
use_vgg=False,
discriminator_blocks=3, discriminator_channels=64,
decoder_loss=1,
encoder_loss=0.7,
adversarial_loss=1e-3,
enable_fp16=args.enable_fp16
)
utils.create_folder_for_run(train_options)
with open(train_options.options_file, 'wb+') as f:
pickle.dump(train_options, f)
pickle.dump(noise_config, f)
pickle.dump(hidden_config, f)
logging.basicConfig(level=logging.INFO,
format='%(message)s',
handlers=[
logging.FileHandler(os.path.join(train_options.this_run_folder, f'{train_options.experiment_name}.log')),
logging.StreamHandler(sys.stdout)
])
logging.info(f'Tensorboard is enabled. Creating logger at {train_options.tb_logger_dir}')
tb_logger = TensorBoardLogger(train_options.tb_logger_dir)
noiser = Noiser(noise_config, device)
model = Hidden(hidden_config, device, noiser, tb_logger)
if args.command == 'continue':
# if we are continuing, we have to load the model params
assert checkpoint is not None
logging.info(f'Loading checkpoint from file {loaded_checkpoint_file_name}')
utils.model_from_checkpoint(model, checkpoint)
logging.info('HiDDeN model: {}\n'.format(model.to_stirng()))
logging.info('Model Configuration:\n')
logging.info(pprint.pformat(vars(hidden_config)))
logging.info('\nNoise configuration:\n')
logging.info(pprint.pformat(str(noise_config)))
logging.info('\nTraining train_options:\n')
logging.info(pprint.pformat(vars(train_options)))
train(model, device, hidden_config, train_options, train_options.this_run_folder, tb_logger)