def build_lr_scheduler(cfg, default_args=None):
if LR_SCHEDULERS.get(cfg['type']):
scheduler = build_from_cfg(cfg, LR_SCHEDULERS, default_args, 'registry')
else:
default_args = dict(optimizer=default_args.get('optimizer'))
scheduler = build_from_cfg(cfg, lr_scheduler, default_args, 'module')
return scheduler
def build_encoder(cfg, default_args=None):
backbone = build_from_cfg(cfg['backbone'], BACKBONES, default_args)
enhance_cfg = cfg.get('enhance')
if enhance_cfg:
enhance_module = build_from_cfg(enhance_cfg, ENHANCE_MODULES,
default_args)
encoder = nn.Sequential(backbone, enhance_module)
else:
encoder = backbone
return encoder
def build_transform(cfgs):
tfs = []
for cfg in cfgs:
if TRANSFORMS.get(cfg['type']):
tf = build_from_cfg(cfg, TRANSFORMS)
else:
tf = build_from_cfg(cfg, albu, mode='module')
tfs.append(tf)
aug = albu.Compose(tfs)
return aug
def build_transform(cfg):
tfs = []
for icfg in cfg:
tf = build_from_cfg(icfg, albu, method='module')
tfs.append(tf)
aug = albu.Compose(tfs)
return aug
def build_transform(cfg):
tfs = []
for icfg in cfg:
tf = build_from_cfg(icfg, TRANSFORMS)
tfs.append(tf)
aug = Compose(tfs)
return aug
def build_module(cfg, default_args=None):
try:
module = build_from_cfg(cfg, UTILS, default_args)
except KeyError as error:
if ' is not in the ' not in error.args[0]:
raise KeyError from error
if ' registry' not in error.args[0]:
raise KeyError from error
module = build_torch_nn(cfg, default_args=default_args)
return module
def build_dataset(cfg, default_args=None):
dataset = build_from_cfg(cfg, DATASETS, default_args)
return dataset
def build_enhance_module(cfg, default_args=None):
#import pdb
#pdb.set_trace()
enhance_module = build_from_cfg(cfg, ENHANCE_MODULES, default_args)
return enhance_module
def __init__(self, cfg):
super().__init__()
#self.criterion = build_from_cfg(cfg, nn, method='module')
self.criterion = build_from_cfg(cfg, CRITERIA, method='registry')
def build_head(cfg, default_args=None):
#import pdb
#pdb.set_trace()
head = build_from_cfg(cfg, HEADS, default_args)
return head
def build_runner(cfg, default_args=None):
runner = build_from_cfg(cfg, RUNNERS, default_args)
return runner
def build_optim(cfg, default_args=None):
optim = build_from_cfg(cfg, torch_optim, default_args, 'module')
return optim
def build_lr_scheduler(cfg, default_args=None):
scheduler = build_from_cfg(cfg, torch_lr_scheduler, default_args, 'module')
return scheduler
def build_decoder(cfg, default_args=None):
decoder = build_from_cfg(cfg, DECODERS, default_args)
return decoder
def build_lr_scheduler(cfg, default_args=None):
scheduler = build_from_cfg(cfg, LR_SCHEDULERS, default_args, 'registry')
return scheduler
def build_enhance_module(cfg, default_args=None):
enhance_module = build_from_cfg(cfg, ENHANCE_MODULES, default_args)
return enhance_module
def build_criterion(cfg):
#criterion = CriterionWrapper(cfg)
criterion = build_from_cfg(cfg, CRITERIA, src='registry')
return criterion
def build_dataloader(cfg, default_args):
loader = build_from_cfg(cfg, torch_data, default_args, 'module')
return loader
def build_torch_nn(cfg, default_args=None):
module = build_from_cfg(cfg, nn, default_args, 'module')
return module
def build_backbone(cfg, default_args=None):
backbone = build_from_cfg(cfg, BACKBONES, default_args)
return backbone
def build_module(cfg, default_args=None):
util = build_from_cfg(cfg, UTILS, default_args)
return util
def build_head(cfg, default_args=None):
head = build_from_cfg(cfg, HEADS, default_args)
return head
def build_brick(cfg, default_args=None):
brick = build_from_cfg(cfg, BRICKS, default_args)
return brick
def build_backbone(cfg, default_args=None):
#import pdb
#pdb.set_trace()
backbone = build_from_cfg(cfg, BACKBONES, default_args)
return backbone
