def parse_args(argv):
"""Parse arguments @argv and return the flags needed for training."""
parser = argparse.ArgumentParser(description=__doc__, allow_abbrev=False)
group = parser.add_argument_group('General Options')
opts.add_general_flags(group)
group = parser.add_argument_group('Dataset Options')
opts.add_dataset_flags(group)
group = parser.add_argument_group('Model Options')
opts.add_model_flags(group)
group = parser.add_argument_group('Soft Label Options')
opts.add_teacher_flags(group)
group = parser.add_argument_group('Training Options')
opts.add_training_flags(group)
update_parser_defaults_from_yaml(parser)
args = parser.parse_args(argv)
# if args.teacher_model is not None and args.teacher_state_file is None:
# parser.error("You should set --teacher-state-file if "
# "--teacher-model is set.")
return args
def main():
# parse command line
parser = prepare_parser()
update_parser_defaults_from_yaml(parser)
config = vars(parser.parse_args())
print(get_dict_str(config))
run(config)
def ddp_init():
"""
use config_cfgnode
"""
from template_lib.v2.config_cfgnode import update_parser_defaults_from_yaml
from template_lib.v2.ddp import parser_local_rank
rank = parser_local_rank()
parser = update_parser_defaults_from_yaml(parser=None,
append_local_rank=True,
is_main_process=(rank == 0))
args = parser.parse_args()
n_gpu = int(os.environ["WORLD_SIZE"]) if "WORLD_SIZE" in os.environ else 1
args.distributed = n_gpu > 1
if args.distributed:
torch.cuda.set_device(args.local_rank)
torch.distributed.init_process_group(backend="nccl",
init_method="env://")
# comm.synchronize()
dist.barrier()
# eval(args.run_func)()
return args
def run():
parser = default_argument_parser()
update_parser_defaults_from_yaml(parser)
args = parser.parse_args()
logger = logging.getLogger('tl')
logger.info(get_dict_str(vars(args)))
launch(
main,
args.num_gpus,
num_machines=args.num_machines,
machine_rank=args.machine_rank,
dist_url=args.dist_url,
args=(args, ),
)
def load_cfg_fom_args(description="Config file options."):
"""Load config from command line arguments and set any specified options."""
parser = argparse.ArgumentParser(description=description)
help_s = "Config file location"
parser.add_argument("--cfg",
dest="cfg_file",
help=help_s,
required=True,
type=str)
help_s = "See pycls/core/config.py for all options"
parser.add_argument("opts",
help=help_s,
default=None,
nargs=argparse.REMAINDER)
if len(sys.argv) == 1:
parser.print_help()
sys.exit(1)
update_parser_defaults_from_yaml(parser)
args = parser.parse_args()
_C.merge_from_file(args.cfg_file)
_C.merge_from_list(args.opts)
def main():
from template_lib.v2.config import update_parser_defaults_from_yaml
parser = argparse.ArgumentParser()
parser.add_argument("--local_rank", type=int, default=0)
parser.add_argument("--run_func", type=str)
update_parser_defaults_from_yaml(parser)
args = parser.parse_args()
n_gpu = int(os.environ["WORLD_SIZE"]) if "WORLD_SIZE" in os.environ else 1
args.distributed = n_gpu > 1
if args.distributed:
torch.cuda.set_device(args.local_rank)
torch.distributed.init_process_group(backend="nccl",
init_method="env://")
# comm.synchronize()
dist.barrier()
# eval(args.run_func)()
return args
def main():
update_parser_defaults_from_yaml(parser)
args = parser.parse_args()
if args.seed is not None:
random.seed(args.seed)
torch.manual_seed(args.seed)
cudnn.deterministic = True
warnings.warn('You have chosen to seed training. '
'This will turn on the CUDNN deterministic setting, '
'which can slow down your training considerably! '
'You may see unexpected behavior when restarting '
'from checkpoints.')
if args.gpu is not None:
warnings.warn('You have chosen a specific GPU. This will completely '
'disable data parallelism.')
if args.dist_url == "env://" and args.world_size == -1:
args.world_size = int(os.environ["WORLD_SIZE"])
args.distributed = args.world_size > 1 or args.multiprocessing_distributed
ngpus_per_node = torch.cuda.device_count()
if args.multiprocessing_distributed:
# Since we have ngpus_per_node processes per node, the total world_size
# needs to be adjusted accordingly
args.world_size = ngpus_per_node * args.world_size
# Use torch.multiprocessing.spawn to launch distributed processes: the
# main_worker process function
mp.spawn(main_worker,
nprocs=ngpus_per_node,
args=(ngpus_per_node, args))
else:
# Simply call main_worker function
main_worker(args.gpu, ngpus_per_node, args)
def get_args():
parser = argparse.ArgumentParser()
# Dataset configuration
parser.add_argument(
'--cGAN',
default=False,
action='store_true',
help='to train cGAN, set this ``True``. default: False')
parser.add_argument(
'--data_root',
type=str,
default='tiny-imagenet-200',
help='path to dataset root directory. default: tiny-imagenet-200')
parser.add_argument('--batch_size',
'-B',
type=int,
default=64,
help='mini-batch size of training data. default: 64')
parser.add_argument(
'--eval_batch_size',
'-eB',
default=None,
help='mini-batch size of evaluation data. default: None')
parser.add_argument(
'--num_workers',
type=int,
default=8,
help='Number of workers for training data loader. default: 8')
# Generator configuration
parser.add_argument(
'--gen_num_features',
'-gnf',
type=int,
default=64,
help=
'Number of features of generator (a.k.a. nplanes or ngf). default: 64')
parser.add_argument(
'--gen_dim_z',
'-gdz',
type=int,
default=128,
help='Dimension of generator input noise. default: 128')
parser.add_argument(
'--gen_bottom_width',
'-gbw',
type=int,
default=4,
help='Initial size of hidden variable of generator. default: 4')
parser.add_argument(
'--gen_distribution',
'-gd',
type=str,
default='normal',
help='Input noise distribution: normal (default) or uniform.')
# Discriminator (Critic) configuration
parser.add_argument(
'--dis_arch_concat',
'-concat',
default=False,
action='store_true',
help='If use concat discriminator, set this true. default: False')
parser.add_argument(
'--dis_emb',
type=int,
default=128,
help='Parameter for concat discriminator. default: 128')
parser.add_argument(
'--dis_num_features',
'-dnf',
type=int,
default=64,
help=
'Number of features of discriminator (a.k.a nplanes or ndf). default: 64'
)
# Optimizer settings
parser.add_argument('--lr',
type=float,
default=0.0002,
help='Initial learning rate of Adam. default: 0.0002')
parser.add_argument('--beta1',
type=float,
default=0.0,
help='beta1 (betas[0]) value of Adam. default: 0.0')
parser.add_argument('--beta2',
type=float,
default=0.9,
help='beta2 (betas[1]) value of Adam. default: 0.9')
parser.add_argument(
'--lr_decay_start',
'-lds',
type=int,
default=50000,
help='Start point of learning rate decay. default: 50000')
# Training setting
parser.add_argument(
'--seed',
type=int,
default=46,
help='Random seed. default: 46 (derived from Nogizaka46)')
parser.add_argument(
'--max_iteration',
'-N',
type=int,
default=100000,
help='Max iteration number of training. default: 100000')
parser.add_argument(
'--n_dis',
type=int,
default=5,
help='Number of discriminator updater per generator updater. default: 5'
)
parser.add_argument(
'--num_classes',
'-nc',
type=int,
default=0,
help='Number of classes in training data. No need to set. default: 0')
parser.add_argument('--loss_type',
type=str,
default='hinge',
help='loss function name. hinge (default) or dcgan.')
parser.add_argument('--relativistic_loss',
'-relloss',
default=False,
action='store_true',
help='Apply relativistic loss or not. default: False')
parser.add_argument(
'--calc_FID',
default=False,
action='store_true',
help='If calculate FID score, set this ``True``. default: False')
# Log and Save interval configuration
parser.add_argument('--results_root',
type=str,
default='results',
help='Path to results directory. default: results')
parser.add_argument(
'--no_tensorboard',
action='store_true',
default=False,
help='If you dislike tensorboard, set this ``False``. default: True')
parser.add_argument(
'--no_image',
action='store_true',
default=False,
help=
'If you dislike saving images on tensorboard, set this ``True``. default: False'
)
parser.add_argument(
'--checkpoint_interval',
'-ci',
type=int,
default=1000,
help=
'Interval of saving checkpoints (model and optimizer). default: 1000')
parser.add_argument('--log_interval',
'-li',
type=int,
default=100,
help='Interval of showing losses. default: 100')
parser.add_argument(
'--eval_interval',
'-ei',
type=int,
default=1000,
help=
'Interval for evaluation (save images and FID calculation). default: 1000'
)
parser.add_argument(
'--n_eval_batches',
'-neb',
type=int,
default=100,
help='Number of mini-batches used in evaluation. default: 100')
parser.add_argument(
'--n_fid_images',
'-nfi',
type=int,
default=5000,
help='Number of images to calculate FID. default: 5000')
parser.add_argument(
'--test',
default=False,
action='store_true',
help='If test this python program, set this ``True``. default: False')
# Resume training
parser.add_argument('--args_path',
default=None,
help='Checkpoint args json path. default: None')
parser.add_argument(
'--gen_ckpt_path',
'-gcp',
default=None,
help='Generator and optimizer checkpoint path. default: None')
parser.add_argument(
'--dis_ckpt_path',
'-dcp',
default=None,
help='Discriminator and optimizer checkpoint path. default: None')
update_parser_defaults_from_yaml(parser)
args = parser.parse_args()
return args
import torch.backends.cudnn as cudnn
from pytorch_lightning import Trainer
from pytorch_lightning.logging import TestTubeLogger
from template_lib.v2.config import update_parser_defaults_from_yaml, global_cfg
from template_lib.utils import update_config
parser = argparse.ArgumentParser(description='Generic runner for VAE models')
parser.add_argument('--config',
'-c',
dest="filename",
metavar='FILE',
help='path to the config file',
default='configs/vae.yaml')
update_parser_defaults_from_yaml(parser)
args = parser.parse_args()
with open(args.filename, 'r') as file:
try:
config = yaml.safe_load(file)
except yaml.YAMLError as exc:
print(exc)
config = update_config(config, global_cfg)
tt_logger = TestTubeLogger(
save_dir=eval(config['logging_params']['save_dir']),
name=config['logging_params']['name'],
debug=False,
create_git_tag=False,
def parse_args():
# Training settings
parser = argparse.ArgumentParser(description='')
parser.add_argument('cmd', choices=['train', 'test'])
parser.add_argument('-d', '--data-dir', default=None, required=True)
parser.add_argument('-l',
'--list-dir',
default=None,
help='List dir to look for train_images.txt etc. '
'It is the same with --data-dir if not set.')
parser.add_argument('-c', '--classes', default=0, type=int)
parser.add_argument('-s', '--crop-size', default=0, type=int)
parser.add_argument('--step', type=int, default=200)
parser.add_argument('--arch')
parser.add_argument('--batch-size',
type=int,
default=64,
metavar='N',
help='input batch size for training (default: 64)')
parser.add_argument('--epochs',
type=int,
default=10,
metavar='N',
help='number of epochs to train (default: 10)')
parser.add_argument('--lr',
type=float,
default=0.01,
metavar='LR',
help='learning rate (default: 0.01)')
parser.add_argument('--lr-mode', type=str, default='step')
parser.add_argument('--momentum',
type=float,
default=0.9,
metavar='M',
help='SGD momentum (default: 0.9)')
parser.add_argument('--weight-decay',
'--wd',
default=1e-4,
type=float,
metavar='W',
help='weight decay (default: 1e-4)')
parser.add_argument('-e',
'--evaluate',
dest='evaluate',
action='store_true',
help='evaluate model on validation set')
parser.add_argument('--resume',
default='',
type=str,
metavar='PATH',
help='path to latest checkpoint (default: none)')
parser.add_argument('--pretrained',
dest='pretrained',
default='',
type=str,
metavar='PATH',
help='use pre-trained model')
parser.add_argument('--save_path',
default='',
type=str,
metavar='PATH',
help='output path for training checkpoints')
parser.add_argument('--save_iter',
default=1,
type=int,
help='number of training iterations between'
'checkpoint history saves')
parser.add_argument('-j', '--workers', type=int, default=8)
parser.add_argument('--load-release', dest='load_rel', default=None)
parser.add_argument('--phase', default='val')
parser.add_argument('--random-scale', default=0, type=float)
parser.add_argument('--random-rotate', default=0, type=int)
parser.add_argument('--bn-sync', action='store_true')
parser.add_argument('--ms',
action='store_true',
help='Turn on multi-scale testing')
parser.add_argument('--with-gt', action='store_true')
parser.add_argument('--test-suffix', default='', type=str)
update_parser_defaults_from_yaml(parser)
args = parser.parse_args()
assert args.classes > 0
print(' '.join(sys.argv))
print(args)
if args.bn_sync:
drn.BatchNorm = batchnormsync.BatchNormSync
return args