Let's eat grass
In the above figure, line 55 contains a mystery. We didn't talk about it in detail earlier. This line is specified, the cfa algorithm, how to train, and the algorithm implementation details are all here.
Then we have to look at the get_model function:
def get_model(config: DictConfig | ListConfig) -> AnomalyModule:
"""Load model from the configuration file.
Works only when the convention for model naming is followed.
The convention for writing model classes is
`anomalib.models.<model_name>.lightning_model.<ModelName>Lightning`
`anomalib.models.stfpm.lightning_model.StfpmLightning`
Args:
config (DictConfig | ListConfig): Config.yaml loaded using OmegaConf
Raises:
ValueError: If unsupported model is passed
Returns:
AnomalyModule: Anomaly Model
"""
logger.info("Loading the model.")
model_list: list[str] = [
"cfa",
"cflow",
"csflow",
"dfkde",
"dfm",
"draem",
"fastflow",
"ganomaly",
"padim",
"patchcore",
"reverse_distillation",
"rkde",
"stfpm",
]
model: AnomalyModule
if config.model.name in model_list:
module = import_module(f"anomalib.models.{config.model.name}")
model = getattr(module, f"{_snake_to_pascal_case(config.model.name)}Lightning")(config)
print("---------------getattr")
print(getattr(module, f"{_snake_to_pascal_case(config.model.name)}Lightning"))
print("---------------getattr-end")
else:
raise ValueError(f"Unknown model {config.model.name}!")
if "init_weights" in config.keys() and config.init_weights:
model.load_state_dict(load(os.path.join(config.project.path, config.init_weights))["state_dict"], strict=False)
return modelHere, the most important thing is,
One, about the module
module = import_module(f"anomalib.models.{config.model.name}")Because in config.yaml, it has been specified,
anomalib.models.{config.model.name} is anomalib.models.cfa
Then, the essence is to execute the following code
module = import_module(anomalib.models.cfa)Remember the above line of code, important!
So, the module is the following thing:
<module 'anomalib.models.cfa' from 'D:\\BaiduNetdiskDownload\\anomalib\\anomalib
-main\\src\\anomalib\\models\\cfa\\__init__.py'>
Second, about the model
model = getattr(module, f"{_snake_to_pascal_case(config.model.name)}Lightning")(config)In fact, it is execution,
model=getattr(module,CfaLightning)(config)
In other words, it is
model=CfaLightning(config)
3. About the training data datamodule
The key is to look at the 54 line
datamodule = get_datamodule(config)
def get_datamodule(config: DictConfig | ListConfig) -> AnomalibDataModule:
"""Get Anomaly Datamodule.
Args:
config (DictConfig | ListConfig): Configuration of the anomaly model.
Returns:
PyTorch Lightning DataModule
"""
logger.info("Loading the datamodule")
datamodule: AnomalibDataModule
# convert center crop to tuple
center_crop = config.dataset.get("center_crop")
if center_crop is not None:
center_crop = (center_crop[0], center_crop[1])
if config.dataset.format.lower() == "mvtec":
datamodule = MVTec(
root=config.dataset.path,
category=config.dataset.category,
image_size=(config.dataset.image_size[0], config.dataset.image_size[1]),
center_crop=center_crop,
normalization=config.dataset.normalization,
train_batch_size=config.dataset.train_batch_size,
eval_batch_size=config.dataset.eval_batch_size,
num_workers=config.dataset.num_workers,
task=config.dataset.task,
transform_config_train=config.dataset.transform_config.train,
transform_config_eval=config.dataset.transform_config.eval,
test_split_mode=config.dataset.test_split_mode,
test_split_ratio=config.dataset.test_split_ratio,
val_split_mode=config.dataset.val_split_mode,
val_split_ratio=config.dataset.val_split_ratio,
)
elif config.dataset.format.lower() == "mvtec_3d":
datamodule = MVTec3D(
root=config.dataset.path,
category=config.dataset.category,
image_size=(config.dataset.image_size[0], config.dataset.image_size[1]),
。。。。。。。。。。
return datamodule
What type of datamodule is this?
AnomalibDataModule type! The parent class of AnomalibDataModule is the LightningDataModule type
class AnomalibDataModule(LightningDataModule, ABC):
"""Base Anomalib data module.
Args:
train_batch_size (int): Batch size used by the train dataloader.
test_batch_size (int): Batch size used by the val and test dataloaders.
num_workers (int): Number of workers used by the train, val and test dataloaders.
test_split_mode (Optional[TestSplitMode], optional): Determines how the test split is obtained.
Options: [none, from_dir, synthetic]
test_split_ratio (float): Fraction of the train images held out for testing.
val_split_mode (ValSplitMode): Determines how the validation split is obtained. Options: [none, same_as_test,
from_test, synthetic]
val_split_ratio (float): Fraction of the train or test images held our for validation.
seed (int | None, optional): Seed used during random subset splitting.
"""
def __init__(
self,
train_batch_size: int,
eval_batch_size: int,
num_workers: int,
val_split_mode: ValSplitMode,
val_split_ratio: float,
test_split_mode: TestSplitMode | None = None,
test_split_ratio: float | None = None,
seed: int | None = None,
) -> None:
super().__init__()
self.train_batch_size = train_batch_size
self.eval_batch_size = eval_batch_size
self.num_workers = num_workers
self.test_split_mode = test_split_mode
self.test_split_ratio = test_split_ratio
self.val_split_mode = val_split_mode
self.val_split_ratio = val_split_ratio
self.seed = seed
self.train_data: AnomalibDataset
self.val_data: AnomalibDataset
self.test_data: AnomalibDataset
self._samples: DataFrame | None = None
def setup(self, stage: str | None = None) -> None:
"""Setup train, validation and test data.
Args:
stage: str | None: Train/Val/Test stages. (Default value = None)
"""
if not self.is_setup:
self._setup(stage)
assert self.is_setup
def _setup(self, _stage: str | None = None) -> None:
"""Set up the datasets and perform dynamic subset splitting.
This method may be overridden in subclass for custom splitting behaviour.
Note: The stage argument is not used here. This is because, for a given instance of an AnomalibDataModule
subclass, all three subsets are created at the first call of setup(). This is to accommodate the subset
splitting behaviour of anomaly tasks, where the validation set is usually extracted from the test set, and
the test set must therefore be created as early as the `fit` stage.
"""
assert self.train_data is not None
assert self.test_data is not None
self.train_data.setup()
self.test_data.setup()
self._create_test_split()
self._create_val_split()
def _create_test_split(self) -> None:
"""Obtain the test set based on the settings in the config."""
if self.test_data.has_normal:
# split the test data into normal and anomalous so these can be processed separately
normal_test_data, self.test_data = split_by_label(self.test_data)
elif self.test_split_mode != TestSplitMode.NONE:
# when the user did not provide any normal images for testing, we sample some from the training set,
# except when the user explicitly requested no test splitting.
logger.info(
"No normal test images found. Sampling from training set using a split ratio of %d",
self.test_split_ratio,
)
if self.test_split_ratio is not None:
self.train_data, normal_test_data = random_split(self.train_data, self.test_split_ratio, seed=self.seed)
if self.test_split_mode == TestSplitMode.FROM_DIR:
self.test_data += normal_test_data
elif self.test_split_mode == TestSplitMode.SYNTHETIC:
self.test_data = SyntheticAnomalyDataset.from_dataset(normal_test_data)
elif self.test_split_mode != TestSplitMode.NONE:
raise ValueError(f"Unsupported Test Split Mode: {self.test_split_mode}")
def _create_val_split(self) -> None:
"""Obtain the validation set based on the settings in the config."""
if self.val_split_mode == ValSplitMode.FROM_TEST:
# randomly sampled from test set
self.test_data, self.val_data = random_split(
self.test_data, self.val_split_ratio, label_aware=True, seed=self.seed
)
elif self.val_split_mode == ValSplitMode.SAME_AS_TEST:
# equal to test set
self.val_data = self.test_data
elif self.val_split_mode == ValSplitMode.SYNTHETIC:
# converted from random training sample
self.train_data, normal_val_data = random_split(self.train_data, self.val_split_ratio, seed=self.seed)
self.val_data = SyntheticAnomalyDataset.from_dataset(normal_val_data)
elif self.val_split_mode != ValSplitMode.NONE:
raise ValueError(f"Unknown validation split mode: {self.val_split_mode}")
@property
def is_setup(self) -> bool:
"""Checks if setup() has been called.
At least one of [train_data, val_data, test_data] should be setup.
"""
_is_setup: bool = False
for data in ("train_data", "val_data", "test_data"):
if hasattr(self, data):
if getattr(self, data).is_setup:
_is_setup = True
return _is_setup
def train_dataloader(self) -> TRAIN_DATALOADERS:
"""Get train dataloader."""
return DataLoader(
dataset=self.train_data, shuffle=True, batch_size=self.train_batch_size, num_workers=self.num_workers
)
def val_dataloader(self) -> EVAL_DATALOADERS:
"""Get validation dataloader."""
return DataLoader(
dataset=self.val_data,
shuffle=False,
batch_size=self.eval_batch_size,
num_workers=self.num_workers,
collate_fn=collate_fn,
)
def test_dataloader(self) -> EVAL_DATALOADERS:
"""Get test dataloader."""
return DataLoader(
dataset=self.test_data,
shuffle=False,
batch_size=self.eval_batch_size,
num_workers=self.num_workers,
collate_fn=collate_fn,
)