def parse_options(is_train=True):
parser = argparse.ArgumentParser()
parser.add_argument('-opt',
type=str,
required=True,
help='Path to option YAML file.')
parser.add_argument('--launcher',
choices=['none', 'pytorch', 'slurm'],
default='none',
help='job launcher')
parser.add_argument('--local_rank', type=int, default=0)
args = parser.parse_args()
opt = parse(args.opt, is_train=is_train)
# distributed settings
if args.launcher == 'none':
opt['dist'] = False
print('Disable distributed.', flush=True)
else:
opt['dist'] = True
if args.launcher == 'slurm' and 'dist_params' in opt:
init_dist(args.launcher, **opt['dist_params'])
else:
init_dist(args.launcher)
opt['rank'], opt['world_size'] = get_dist_info()
# random seed
seed = opt.get('manual_seed')
if seed is None:
seed = random.randint(1, 10000)
opt['manual_seed'] = seed
set_random_seed(seed + opt['rank'])
return opt
def parse_options(root_path, is_train=True):
parser = argparse.ArgumentParser()
parser.add_argument('-opt',
type=str,
required=True,
help='Path to option YAML file.')
parser.add_argument('--launcher',
choices=['none', 'pytorch', 'slurm'],
default='none',
help='job launcher')
parser.add_argument('--auto_resume', action='store_true')
parser.add_argument('--debug', action='store_true')
parser.add_argument('--local_rank', type=int, default=0)
parser.add_argument(
'--force_yml',
nargs='+',
default=None,
help='Force to update yml files. Examples: train:ema_decay=0.999')
args = parser.parse_args()
# parse yml to dict
with open(args.opt, mode='r') as f:
opt = yaml.load(f, Loader=ordered_yaml()[0])
# distributed settings
if args.launcher == 'none':
opt['dist'] = False
print('Disable distributed.', flush=True)
else:
opt['dist'] = True
if args.launcher == 'slurm' and 'dist_params' in opt:
init_dist(args.launcher, **opt['dist_params'])
else:
init_dist(args.launcher)
opt['rank'], opt['world_size'] = get_dist_info()
# random seed
seed = opt.get('manual_seed')
if seed is None:
seed = random.randint(1, 10000)
opt['manual_seed'] = seed
set_random_seed(seed + opt['rank'])
# force to update yml options
if args.force_yml is not None:
for entry in args.force_yml:
# now do not support creating new keys
keys, value = entry.split('=')
keys, value = keys.strip(), value.strip()
value = _postprocess_yml_value(value)
eval_str = 'opt'
for key in keys.split(':'):
eval_str += f'["{key}"]'
eval_str += '=value'
# using exec function
exec(eval_str)
opt['auto_resume'] = args.auto_resume
opt['is_train'] = is_train
# debug setting
if args.debug and not opt['name'].startswith('debug'):
opt['name'] = 'debug_' + opt['name']
if opt['num_gpu'] == 'auto':
opt['num_gpu'] = torch.cuda.device_count()
# datasets
for phase, dataset in opt['datasets'].items():
# for multiple datasets, e.g., val_1, val_2; test_1, test_2
phase = phase.split('_')[0]
dataset['phase'] = phase
if 'scale' in opt:
dataset['scale'] = opt['scale']
if dataset.get('dataroot_gt') is not None:
dataset['dataroot_gt'] = osp.expanduser(dataset['dataroot_gt'])
if dataset.get('dataroot_lq') is not None:
dataset['dataroot_lq'] = osp.expanduser(dataset['dataroot_lq'])
# paths
for key, val in opt['path'].items():
if (val is not None) and ('resume_state' in key
or 'pretrain_network' in key):
opt['path'][key] = osp.expanduser(val)
if is_train:
experiments_root = osp.join(root_path, 'experiments', opt['name'])
opt['path']['experiments_root'] = experiments_root
opt['path']['models'] = osp.join(experiments_root, 'models')
opt['path']['training_states'] = osp.join(experiments_root,
'training_states')
opt['path']['log'] = experiments_root
opt['path']['visualization'] = osp.join(experiments_root,
'visualization')
# change some options for debug mode
if 'debug' in opt['name']:
if 'val' in opt:
opt['val']['val_freq'] = 8
opt['logger']['print_freq'] = 1
opt['logger']['save_checkpoint_freq'] = 8
else: # test
results_root = osp.join(root_path, 'results', opt['name'])
opt['path']['results_root'] = results_root
opt['path']['log'] = results_root
opt['path']['visualization'] = osp.join(results_root, 'visualization')
return opt, args