def build(config):
if 'transforms' in config:
config['transforms'] = [build(transform) for transform in config['transforms']]
try:
return object_from_dict(config, A)
except AttributeError:
return object_from_dict(config, pytorch)
def build(config):
if "transforms" in config:
config["transforms"] = [
build(transform) for transform in config["transforms"]
]
if hasattr(A, config["type"]):
return object_from_dict(config, A)
elif hasattr(pytorch, config["type"]):
return object_from_dict(config, pytorch)
else:
return object_from_dict(config, ALBUMENTATIONS)
def build(config: ConfigDict) -> T:
if "transforms" in config:
config["transforms"] = [
build(transform) for transform in config["transforms"]
]
try:
return object_from_dict(config, A)
except AttributeError:
try:
return object_from_dict(config, pytorch)
except AttributeError:
return object_from_dict(config)
def metrics(self) -> Optional[Dict[str, T]]:
if hasattr(self.config, "METRICS"):
return {
k: object_from_dict(v)
for k, v in self.config.METRICS.items()
}
return None
def dataset(self, mode: str) -> Dataset:
arguments = self.config.DATA[mode]
arguments.update({
"image_transforms":
self.transform(mode, level="image_transforms")
})
arguments.update(
{"crop_transforms": self.transform(mode, level="crop_transforms")})
return object_from_dict(arguments)
def losses(self) -> Dict[str, T]:
losses = {
k: object_from_dict(v)
for k, v in self.config.LOSSES.items()
}
if self._has_bdp_hook:
return {
k: BalancedDataParallelCriterion(v)
for k, v in losses.items()
}
return losses
def make_model(
config: ConfigDict, device: torch.device = torch.device("cpu")
) -> torch.nn.Module:
model = object_from_dict(config)
return model.to(device)