Datamodules

terratorch.datamodules.biomassters

BioMasstersNonGeoDataModule

Bases: NonGeoDataModule

NonGeo LightningDataModule implementation for BioMassters datamodule.

__init__(data_root, batch_size=4, num_workers=0, bands=BioMasstersNonGeo.all_band_names, train_transform=None, val_transform=None, test_transform=None, predict_transform=None, aug=None, drop_last=True, sensors=['S1', 'S2'], as_time_series=False, metadata_filename=default_metadata_filename, max_cloud_percentage=None, max_red_mean=None, include_corrupt=True, subset=1, seed=42, use_four_frames=False, **kwargs)

Initializes the DataModule for the non-geospatial BioMassters datamodule.

Parameters:

Name	Type	Description	Default
data_root	str	Root directory containing the dataset.	required
batch_size	int	Batch size for DataLoaders. Defaults to 4.	4
num_workers	int	Number of workers for data loading. Defaults to 0.	0
bands	dict[str, Sequence[str]] | Sequence[str]	Band configuration; either a dict mapping sensors to bands or a list for the first sensor. Defaults to BioMasstersNonGeo.all_band_names	all_band_names
train_transform	Compose | None | list[BasicTransform]	Transformations for training data.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation data.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing data.	None
predict_transform	Compose | None | list[BasicTransform]	Transformations for prediction data.	None
aug	AugmentationSequential	Augmentation or normalization to apply. Defaults to normalization if not provided.	None
drop_last	bool	Whether to drop the last incomplete batch. Defaults to True.	True
sensors	Sequence[str]	List of sensors to use (e.g., ["S1", "S2"]). Defaults to ["S1", "S2"].	['S1', 'S2']
as_time_series	bool	Whether to treat data as a time series. Defaults to False.	False
metadata_filename	str	Metadata filename. Defaults to "The_BioMassters_-_features_metadata.csv.csv".	default_metadata_filename
max_cloud_percentage	float | None	Maximum allowed cloud percentage. Defaults to None.	None
max_red_mean	float | None	Maximum allowed red band mean. Defaults to None.	None
include_corrupt	bool	Whether to include corrupt data. Defaults to True.	True
subset	float	Fraction of the dataset to use. Defaults to 1.	1
seed	int	Random seed for reproducibility. Defaults to 42.	42
use_four_frames	bool	Whether to use a four frames configuration. Defaults to False.	False
**kwargs	Any	Additional keyword arguments.	{}

Returns:

Type	Description
None	None.

setup(stage)

Set up datasets.

Parameters:

Name	Type	Description	Default
stage	str	Either fit, validate, test, or predict.	required

terratorch.datamodules.burn_intensity

BurnIntensityNonGeoDataModule

Bases: NonGeoDataModule

NonGeo LightningDataModule implementation for BurnIntensity datamodule.

__init__(data_root, batch_size=4, num_workers=0, bands=BurnIntensityNonGeo.all_band_names, train_transform=None, val_transform=None, test_transform=None, predict_transform=None, use_full_data=True, no_data_replace=0.0001, no_label_replace=-1, use_metadata=False, **kwargs)

Initializes the DataModule for the BurnIntensity non-geospatial datamodule.

Parameters:

Name	Type	Description	Default
data_root	str	Root directory of the dataset.	required
batch_size	int	Batch size for DataLoaders. Defaults to 4.	4
num_workers	int	Number of workers for data loading. Defaults to 0.	0
bands	Sequence[str]	List of bands to use. Defaults to BurnIntensityNonGeo.all_band_names.	all_band_names
train_transform	Compose | None | list[BasicTransform]	Transformations for training.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing.	None
predict_transform	Compose | None | list[BasicTransform]	Transformations for prediction.	None
use_full_data	bool	Whether to use the full dataset or data with less than 25 percent zeros. Defaults to True.	True
no_data_replace	float | None	Value to replace missing data. Defaults to 0.0001.	0.0001
no_label_replace	int | None	Value to replace missing labels. Defaults to -1.	-1
use_metadata	bool	Whether to return metadata info (time and location).	False
**kwargs	Any	Additional keyword arguments.	{}

setup(stage)

Set up datasets.

Parameters:

Name	Type	Description	Default
stage	str	Either fit, validate, test, or predict.	required

terratorch.datamodules.carbonflux

CarbonFluxNonGeoDataModule

Bases: NonGeoDataModule

NonGeo LightningDataModule implementation for Carbon FLux dataset.

__init__(data_root, batch_size=4, num_workers=0, bands=CarbonFluxNonGeo.all_band_names, train_transform=None, val_transform=None, test_transform=None, predict_transform=None, aug=None, no_data_replace=0.0001, use_metadata=False, **kwargs)

Initializes the CarbonFluxNonGeoDataModule.

Parameters:

Name	Type	Description	Default
data_root	str	Root directory of the dataset.	required
batch_size	int	Batch size for DataLoaders. Defaults to 4.	4
num_workers	int	Number of workers for data loading. Defaults to 0.	0
bands	Sequence[str]	List of bands to use. Defaults to CarbonFluxNonGeo.all_band_names.	all_band_names
train_transform	Compose | None | list[BasicTransform]	Transformations for training data.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation data.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing data.	None
predict_transform	Compose | None | list[BasicTransform]	Transformations for prediction data.	None
aug	AugmentationSequential	Augmentation sequence; if None, applies multimodal normalization.	None
no_data_replace	float | None	Value to replace missing data. Defaults to 0.0001.	0.0001
use_metadata	bool	Whether to return metadata info.	False
**kwargs	Any	Additional keyword arguments.	{}

setup(stage)

Set up datasets.

Parameters:

Name	Type	Description	Default
stage	str	Either fit, validate, test, or predict.	required

terratorch.datamodules.forestnet

ForestNetNonGeoDataModule

Bases: NonGeoDataModule

NonGeo LightningDataModule implementation for Landslide4Sense dataset.

__init__(data_root, batch_size=4, num_workers=0, label_map=ForestNetNonGeo.default_label_map, bands=ForestNetNonGeo.all_band_names, train_transform=None, val_transform=None, test_transform=None, predict_transform=None, fraction=1.0, aug=None, use_metadata=False, **kwargs)

Initializes the ForestNetNonGeoDataModule.

Parameters:

Name	Type	Description	Default
data_root	str	Directory containing the dataset.	required
batch_size	int	Batch size for data loaders. Defaults to 4.	4
num_workers	int	Number of workers for data loading. Defaults to 0.	0
label_map	dict[str, int]	Mapping of labels to integers. Defaults to ForestNetNonGeo.default_label_map.	default_label_map
bands	Sequence[str]	List of band names to use. Defaults to ForestNetNonGeo.all_band_names.	all_band_names
train_transform	Compose | None | list[BasicTransform]	Transformations for training data.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation data.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing data.	None
predict_transform	Compose | None | list[BasicTransform]	Transformations for prediction.	None
fraction	float	Fraction of data to use. Defaults to 1.0.	1.0
aug	AugmentationSequential	Augmentation/normalization pipeline; if None, uses Normalize.	None
use_metadata	bool	Whether to return metadata info.	False
**kwargs	Any	Additional keyword arguments.	{}

setup(stage)

Set up datasets.

Parameters:

Name	Type	Description	Default
stage	str	Either fit, validate, test, or predict.	required

terratorch.datamodules.fire_scars

FireScarsDataModule

Bases: GeoDataModule

Geo Fire Scars data module implementation that merges input data with ground truth segmentation masks.

FireScarsNonGeoDataModule

Bases: NonGeoDataModule

NonGeo LightningDataModule implementation for Fire Scars dataset.

__init__(data_root, batch_size=4, num_workers=0, bands=FireScarsNonGeo.all_band_names, train_transform=None, val_transform=None, test_transform=None, predict_transform=None, drop_last=True, no_data_replace=0, no_label_replace=-1, use_metadata=False, **kwargs)

Initializes the FireScarsNonGeoDataModule.

Parameters:

Name	Type	Description	Default
data_root	str	Root directory of the dataset.	required
batch_size	int	Batch size for DataLoaders. Defaults to 4.	4
num_workers	int	Number of workers for data loading. Defaults to 0.	0
bands	Sequence[str]	List of band names. Defaults to FireScarsNonGeo.all_band_names.	all_band_names
train_transform	Compose | None | list[BasicTransform]	Transformations for training.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing.	None
predict_transform	Compose | None | list[BasicTransform]	Transformations for prediction.	None
drop_last	bool	Whether to drop the last incomplete batch. Defaults to True.	True
no_data_replace	float | None	Replacement value for missing data. Defaults to 0.	0
no_label_replace	int | None	Replacement value for missing labels. Defaults to -1.	-1
use_metadata	bool	Whether to return metadata info.	False
**kwargs	Any	Additional keyword arguments.	{}

setup(stage)

Set up datasets.

Parameters:

Name	Type	Description	Default
stage	str	Either fit, validate, test, or predict.	required

terratorch.datamodules.landslide4sense

Landslide4SenseNonGeoDataModule

Bases: NonGeoDataModule

NonGeo LightningDataModule implementation for Landslide4Sense dataset.

__init__(data_root, batch_size=4, num_workers=0, bands=Landslide4SenseNonGeo.all_band_names, train_transform=None, val_transform=None, test_transform=None, predict_transform=None, aug=None, **kwargs)

Initializes the Landslide4SenseNonGeoDataModule.

Parameters:

Name	Type	Description	Default
data_root	str	Root directory of the dataset.	required
batch_size	int	Batch size for data loaders. Defaults to 4.	4
num_workers	int	Number of workers for data loading. Defaults to 0.	0
bands	Sequence[str]	List of band names to use. Defaults to Landslide4SenseNonGeo.all_band_names.	all_band_names
train_transform	Compose | None | list[BasicTransform]	Transformations for training data.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation data.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing data.	None
predict_transform	Compose | None | list[BasicTransform]	Transformations for prediction data.	None
aug	AugmentationSequential	Augmentation pipeline; if None, applies normalization using computed means and stds.	None
**kwargs	Any	Additional keyword arguments.	{}

setup(stage)

Set up datasets.

Parameters:

Name	Type	Description	Default
stage	str	Either fit, validate, test, or predict.	required

terratorch.datamodules.m_eurosat

MEuroSATNonGeoDataModule

Bases: GeobenchDataModule

NonGeo LightningDataModule implementation for M-EuroSAT dataset.

__init__(batch_size=8, num_workers=0, data_root='./', bands=None, train_transform=None, val_transform=None, test_transform=None, aug=None, partition='default', **kwargs)

Initializes the MEuroSATNonGeoDataModule for the MEuroSATNonGeo dataset.

Parameters:

Name	Type	Description	Default
batch_size	int	Batch size for DataLoaders. Defaults to 8.	8
num_workers	int	Number of workers for data loading. Defaults to 0.	0
data_root	str	Root directory of the dataset. Defaults to "./".	'./'
bands	Sequence[str] | None	List of bands to use. Defaults to None.	None
train_transform	Compose | None | list[BasicTransform]	Transformations for training.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing.	None
aug	AugmentationSequential	Augmentation/normalization pipeline. Defaults to None.	None
partition	str	Partition size. Defaults to "default".	'default'
**kwargs	Any	Additional keyword arguments.	{}

terratorch.datamodules.m_bigearthnet

MBigEarthNonGeoDataModule

Bases: GeobenchDataModule

NonGeo LightningDataModule implementation for M-BigEarthNet dataset.

__init__(batch_size=8, num_workers=0, data_root='./', bands=None, train_transform=None, val_transform=None, test_transform=None, aug=None, partition='default', **kwargs)

Initializes the MBigEarthNonGeoDataModule for the M-BigEarthNet dataset.

Parameters:

Name	Type	Description	Default
batch_size	int	Batch size for DataLoaders. Defaults to 8.	8
num_workers	int	Number of workers for data loading. Defaults to 0.	0
data_root	str	Root directory of the dataset. Defaults to "./".	'./'
bands	Sequence[str] | None	List of bands to use. Defaults to None.	None
train_transform	Compose | None | list[BasicTransform]	Transformations for training.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing.	None
aug	AugmentationSequential	Augmentation/normalization pipeline. Defaults to None.	None
partition	str	Partition size. Defaults to "default".	'default'
**kwargs	Any	Additional keyword arguments.	{}

terratorch.datamodules.m_brick_kiln

MBrickKilnNonGeoDataModule

Bases: GeobenchDataModule

NonGeo LightningDataModule implementation for M-BrickKiln dataset.

__init__(batch_size=8, num_workers=0, data_root='./', bands=None, train_transform=None, val_transform=None, test_transform=None, aug=None, partition='default', **kwargs)

Initializes the MBrickKilnNonGeoDataModule for the M-BrickKilnNonGeo dataset.

Parameters:

Name	Type	Description	Default
batch_size	int	Batch size for DataLoaders. Defaults to 8.	8
num_workers	int	Number of workers for data loading. Defaults to 0.	0
data_root	str	Root directory of the dataset. Defaults to "./".	'./'
bands	Sequence[str] | None	List of bands to use. Defaults to None.	None
train_transform	Compose | None | list[BasicTransform]	Transformations for training.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing.	None
aug	AugmentationSequential	Augmentation/normalization pipeline. Defaults to None.	None
partition	str	Partition size. Defaults to "default".	'default'
**kwargs	Any	Additional keyword arguments.	{}

terratorch.datamodules.m_forestnet

MForestNetNonGeoDataModule

Bases: GeobenchDataModule

NonGeo LightningDataModule implementation for M-ForestNet dataset.

__init__(batch_size=8, num_workers=0, data_root='./', bands=None, train_transform=None, val_transform=None, test_transform=None, aug=None, partition='default', use_metadata=False, **kwargs)

Initializes the MForestNetNonGeoDataModule for the MForestNetNonGeo dataset.

Parameters:

Name	Type	Description	Default
batch_size	int	Batch size for DataLoaders. Defaults to 8.	8
num_workers	int	Number of workers for data loading. Defaults to 0.	0
data_root	str	Root directory of the dataset. Defaults to "./".	'./'
bands	Sequence[str] | None	List of bands to use. Defaults to None.	None
train_transform	Compose | None | list[BasicTransform]	Transformations for training.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing.	None
aug	AugmentationSequential	Augmentation/normalization pipeline. Defaults to None.	None
partition	str	Partition size. Defaults to "default".	'default'
use_metadata	bool	Whether to return metadata info.	False
**kwargs	Any	Additional keyword arguments.	{}

terratorch.datamodules.m_so2sat

MSo2SatNonGeoDataModule

Bases: GeobenchDataModule

NonGeo LightningDataModule implementation for M-So2Sat dataset.

__init__(batch_size=8, num_workers=0, data_root='./', bands=None, train_transform=None, val_transform=None, test_transform=None, aug=None, partition='default', **kwargs)

Initializes the MSo2SatNonGeoDataModule for the MSo2SatNonGeo dataset.

Parameters:

Name	Type	Description	Default
batch_size	int	Batch size for DataLoaders. Defaults to 8.	8
num_workers	int	Number of workers for data loading. Defaults to 0.	0
data_root	str	Root directory of the dataset. Defaults to "./".	'./'
bands	Sequence[str] | None	List of bands to use. Defaults to None.	None
train_transform	Compose | None | list[BasicTransform]	Transformations for training.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing.	None
aug	AugmentationSequential	Augmentation/normalization pipeline. Defaults to None.	None
partition	str	Partition size. Defaults to "default".	'default'
**kwargs	Any	Additional keyword arguments.	{}

terratorch.datamodules.m_pv4ger

MPv4gerNonGeoDataModule

Bases: GeobenchDataModule

NonGeo LightningDataModule implementation for M-Pv4ger dataset.

__init__(batch_size=8, num_workers=0, data_root='./', bands=None, train_transform=None, val_transform=None, test_transform=None, aug=None, partition='default', use_metadata=False, **kwargs)

Initializes the MPv4gerNonGeoDataModule for the MPv4gerNonGeo dataset.

Parameters:

Name	Type	Description	Default
batch_size	int	Batch size for DataLoaders. Defaults to 8.	8
num_workers	int	Number of workers for data loading. Defaults to 0.	0
data_root	str	Root directory of the dataset. Defaults to "./".	'./'
bands	Sequence[str] | None	List of bands to use. Defaults to None.	None
train_transform	Compose | None | list[BasicTransform]	Transformations for training.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing.	None
aug	AugmentationSequential	Augmentation/normalization pipeline. Defaults to None.	None
partition	str	Partition size. Defaults to "default".	'default'
use_metadata	bool	Whether to return metadata info.	False
**kwargs	Any	Additional keyword arguments.	{}

terratorch.datamodules.m_cashew_plantation

MBeninSmallHolderCashewsNonGeoDataModule

Bases: GeobenchDataModule

NonGeo LightningDataModule implementation for M-Cashew Plantation dataset.

__init__(batch_size=8, num_workers=0, data_root='./', bands=None, train_transform=None, val_transform=None, test_transform=None, aug=None, partition='default', use_metadata=False, **kwargs)

Initializes the MBeninSmallHolderCashewsNonGeoDataModule for the M-BeninSmallHolderCashewsNonGeo dataset.

Parameters:

Name	Type	Description	Default
batch_size	int	Batch size for DataLoaders. Defaults to 8.	8
num_workers	int	Number of workers for data loading. Defaults to 0.	0
data_root	str	Root directory of the dataset. Defaults to "./".	'./'
bands	Sequence[str] | None	List of bands to use. Defaults to None.	None
train_transform	Compose | None | list[BasicTransform]	Transformations for training.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing.	None
aug	AugmentationSequential	Augmentation/normalization pipeline. Defaults to None.	None
partition	str	Partition size. Defaults to "default".	'default'
use_metadata	bool	Whether to return metadata info.	False
**kwargs	Any	Additional keyword arguments.	{}

terratorch.datamodules.m_nz_cattle

MNzCattleNonGeoDataModule

Bases: GeobenchDataModule

NonGeo LightningDataModule implementation for M-NZCattle dataset.

__init__(batch_size=8, num_workers=0, data_root='./', bands=None, train_transform=None, val_transform=None, test_transform=None, aug=None, partition='default', use_metadata=False, **kwargs)

Initializes the MNzCattleNonGeoDataModule for the MNzCattleNonGeo dataset.

Parameters:

Name	Type	Description	Default
batch_size	int	Batch size for DataLoaders. Defaults to 8.	8
num_workers	int	Number of workers for data loading. Defaults to 0.	0
data_root	str	Root directory of the dataset. Defaults to "./".	'./'
bands	Sequence[str] | None	List of bands to use. Defaults to None.	None
train_transform	Compose | None | list[BasicTransform]	Transformations for training.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing.	None
aug	AugmentationSequential	Augmentation/normalization pipeline. Defaults to None.	None
partition	str	Partition size. Defaults to "default".	'default'
use_metadata	bool	Whether to return metadata info.	False
**kwargs	Any	Additional keyword arguments.	{}

terratorch.datamodules.m_chesapeake_landcover

MChesapeakeLandcoverNonGeoDataModule

Bases: GeobenchDataModule

NonGeo LightningDataModule implementation for M-ChesapeakeLandcover dataset.

__init__(batch_size=8, num_workers=0, data_root='./', bands=None, train_transform=None, val_transform=None, test_transform=None, aug=None, partition='default', **kwargs)

Initializes the MChesapeakeLandcoverNonGeoDataModule for the M-BigEarthNet dataset.

Parameters:

Name	Type	Description	Default
batch_size	int	Batch size for DataLoaders. Defaults to 8.	8
num_workers	int	Number of workers for data loading. Defaults to 0.	0
data_root	str	Root directory of the dataset. Defaults to "./".	'./'
bands	Sequence[str] | None	List of bands to use. Defaults to None.	None
train_transform	Compose | None | list[BasicTransform]	Transformations for training.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing.	None
aug	AugmentationSequential	Augmentation/normalization pipeline. Defaults to None.	None
partition	str	Partition size. Defaults to "default".	'default'
**kwargs	Any	Additional keyword arguments.	{}

terratorch.datamodules.m_pv4ger_seg

MPv4gerSegNonGeoDataModule

Bases: GeobenchDataModule

NonGeo LightningDataModule implementation for M-Pv4gerSeg dataset.

__init__(batch_size=8, num_workers=0, data_root='./', bands=None, train_transform=None, val_transform=None, test_transform=None, aug=None, partition='default', use_metadata=False, **kwargs)

Initializes the MPv4gerNonGeoDataModule for the MPv4gerSegNonGeo dataset.

Parameters:

Name	Type	Description	Default
batch_size	int	Batch size for DataLoaders. Defaults to 8.	8
num_workers	int	Number of workers for data loading. Defaults to 0.	0
data_root	str	Root directory of the dataset. Defaults to "./".	'./'
bands	Sequence[str] | None	List of bands to use. Defaults to None.	None
train_transform	Compose | None | list[BasicTransform]	Transformations for training.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing.	None
aug	AugmentationSequential	Augmentation/normalization pipeline. Defaults to None.	None
partition	str	Partition size. Defaults to "default".	'default'
use_metadata	bool	Whether to return metadata info.	False
**kwargs	Any	Additional keyword arguments.	{}

terratorch.datamodules.m_SA_crop_type

MSACropTypeNonGeoDataModule

Bases: GeobenchDataModule

NonGeo LightningDataModule implementation for M-SA-CropType dataset.

__init__(batch_size=8, num_workers=0, data_root='./', bands=None, train_transform=None, val_transform=None, test_transform=None, aug=None, partition='default', **kwargs)

Initializes the MSACropTypeNonGeoDataModule for the MSACropTypeNonGeo dataset.

Parameters:

Name	Type	Description	Default
batch_size	int	Batch size for DataLoaders. Defaults to 8.	8
num_workers	int	Number of workers for data loading. Defaults to 0.	0
data_root	str	Root directory of the dataset. Defaults to "./".	'./'
bands	Sequence[str] | None	List of bands to use. Defaults to None.	None
train_transform	Compose | None | list[BasicTransform]	Transformations for training.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing.	None
aug	AugmentationSequential	Augmentation/normalization pipeline. Defaults to None.	None
partition	str	Partition size. Defaults to "default".	'default'
**kwargs	Any	Additional keyword arguments.	{}

terratorch.datamodules.m_neontree

MNeonTreeNonGeoDataModule

Bases: GeobenchDataModule

NonGeo LightningDataModule implementation for M-NeonTree dataset.

__init__(batch_size=8, num_workers=0, data_root='./', bands=None, train_transform=None, val_transform=None, test_transform=None, aug=None, partition='default', **kwargs)

Initializes the MNeonTreeNonGeoDataModule for the MNeonTreeNonGeo dataset.

Parameters:

Name	Type	Description	Default
batch_size	int	Batch size for DataLoaders. Defaults to 8.	8
num_workers	int	Number of workers for data loading. Defaults to 0.	0
data_root	str	Root directory of the dataset. Defaults to "./".	'./'
bands	Sequence[str] | None	List of bands to use. Defaults to None.	None
train_transform	Compose | None | list[BasicTransform]	Transformations for training.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing.	None
aug	AugmentationSequential	Augmentation/normalization pipeline. Defaults to None.	None
partition	str	Partition size. Defaults to "default".	'default'
**kwargs	Any	Additional keyword arguments.	{}

terratorch.datamodules.multi_temporal_crop_classification

MultiTemporalCropClassificationDataModule

Bases: NonGeoDataModule

NonGeo LightningDataModule implementation for multi-temporal crop classification.

__init__(data_root, batch_size=4, num_workers=0, bands=MultiTemporalCropClassification.all_band_names, train_transform=None, val_transform=None, test_transform=None, predict_transform=None, drop_last=True, no_data_replace=0, no_label_replace=-1, expand_temporal_dimension=True, reduce_zero_label=True, use_metadata=False, metadata_file_name='chips_df.csv', **kwargs)

Initializes the MultiTemporalCropClassificationDataModule for multi-temporal crop classification.

Parameters:

Name	Type	Description	Default
data_root	str	Directory containing the dataset.	required
batch_size	int	Batch size for DataLoaders. Defaults to 4.	4
num_workers	int	Number of workers for data loading. Defaults to 0.	0
bands	Sequence[str]	List of bands to use. Defaults to MultiTemporalCropClassification.all_band_names.	all_band_names
train_transform	Compose | None | list[BasicTransform]	Transformations for training data.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation data.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing data.	None
predict_transform	Compose | None | list[BasicTransform]	Transformations for prediction data.	None
drop_last	bool	Whether to drop the last incomplete batch during training. Defaults to True.	True
no_data_replace	float | None	Replacement value for missing data. Defaults to 0.	0
no_label_replace	int | None	Replacement value for missing labels. Defaults to -1.	-1
expand_temporal_dimension	bool	Go from shape (time*channels, h, w) to (channels, time, h, w). Defaults to True.	True
reduce_zero_label	bool	Subtract 1 from all labels. Useful when labels start from 1 instead of the expected 0. Defaults to True.	True
use_metadata	bool	Whether to return metadata info (time and location).	False
**kwargs	Any	Additional keyword arguments.	{}

setup(stage)

Set up datasets.

Parameters:

Name	Type	Description	Default
stage	str	Either fit, validate, test, or predict.	required

terratorch.datamodules.open_sentinel_map

OpenSentinelMapDataModule

Bases: NonGeoDataModule

NonGeo LightningDataModule implementation for Open Sentinel Map.

__init__(bands=None, batch_size=8, num_workers=0, data_root='./', train_transform=None, val_transform=None, test_transform=None, predict_transform=None, spatial_interpolate_and_stack_temporally=True, pad_image=None, truncate_image=None, **kwargs)

Initializes the OpenSentinelMapDataModule for the Open Sentinel Map dataset.

Parameters:

Name	Type	Description	Default
bands	list[str] | None	List of bands to use. Defaults to None.	None
batch_size	int	Batch size for DataLoaders. Defaults to 8.	8
num_workers	int	Number of workers for data loading. Defaults to 0.	0
data_root	str	Root directory of the dataset. Defaults to "./".	'./'
train_transform	Compose | None | list[BasicTransform]	Transformations for training data.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation data.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing data.	None
predict_transform	Compose | None | list[BasicTransform]	Transformations for prediction data.	None
spatial_interpolate_and_stack_temporally	bool	If True, the bands are interpolated and concatenated over time. Default is True.	True
pad_image	int | None	Number of timesteps to pad the time dimension of the image. If None, no padding is applied.	None
truncate_image	int | None	Number of timesteps to truncate the time dimension of the image. If None, no truncation is performed.	None
**kwargs	Any	Additional keyword arguments.	{}

setup(stage)

Set up datasets.

Parameters:

Name	Type	Description	Default
stage	str	Either fit, validate, test, or predict.	required

terratorch.datamodules.openearthmap

OpenEarthMapNonGeoDataModule

Bases: NonGeoDataModule

NonGeo LightningDataModule implementation for Open Earth Map.

__init__(batch_size=8, num_workers=0, data_root='./', train_transform=None, val_transform=None, test_transform=None, predict_transform=None, aug=None, **kwargs)

Initializes the OpenEarthMapNonGeoDataModule for the Open Earth Map dataset.

Parameters:

Name	Type	Description	Default
batch_size	int	Batch size for DataLoaders. Defaults to 8.	8
num_workers	int	Number of workers for data loading. Defaults to 0.	0
data_root	str	Root directory of the dataset. Defaults to "./".	'./'
train_transform	Compose | None | list[BasicTransform]	Transformations for training data.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation data.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for test data.	None
predict_transform	Compose | None | list[BasicTransform]	Transformations for prediction data.	None
aug	AugmentationSequential	Augmentation pipeline; if None, defaults to normalization using computed means and stds.	None
**kwargs	Any	Additional keyword arguments. Can include 'bands' (list[str]) to specify the bands; defaults to OpenEarthMapNonGeo.all_band_names if not provided.	{}

setup(stage)

Set up datasets.

Parameters:

Name	Type	Description	Default
stage	str	Either fit, validate, test, or predict.	required

terratorch.datamodules.pastis

PASTISDataModule

Bases: NonGeoDataModule

NonGeo LightningDataModule implementation for PASTIS.

__init__(batch_size=8, num_workers=0, data_root='./', truncate_image=None, pad_image=None, train_transform=None, val_transform=None, test_transform=None, predict_transform=None, **kwargs)

Initializes the PASTISDataModule for the PASTIS dataset.

Parameters:

Name	Type	Description	Default
batch_size	int	Batch size for DataLoaders. Defaults to 8.	8
num_workers	int	Number of workers for data loading. Defaults to 0.	0
data_root	str	Directory containing the dataset. Defaults to "./".	'./'
truncate_image	int	Truncate the time dimension of the image to a specified number of timesteps. If None, no truncation is performed.	None
pad_image	int	Pad the time dimension of the image to a specified number of timesteps. If None, no padding is applied.	None
train_transform	Compose | None | list[BasicTransform]	Transformations for training data.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation data.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for testing data.	None
predict_transform	Compose | None | list[BasicTransform]	Transformations for prediction data.	None
**kwargs	Any	Additional keyword arguments.	{}

setup(stage)

Set up datasets.

Parameters:

Name	Type	Description	Default
stage	str	Either fit, validate, test, or predict.	required

terratorch.datamodules.sen1floods11

Sen1Floods11NonGeoDataModule

Bases: NonGeoDataModule

NonGeo LightningDataModule implementation for Fire Scars.

__init__(data_root, batch_size=4, num_workers=0, bands=Sen1Floods11NonGeo.all_band_names, train_transform=None, val_transform=None, test_transform=None, predict_transform=None, drop_last=True, constant_scale=0.0001, no_data_replace=0, no_label_replace=-1, use_metadata=False, **kwargs)

Initializes the Sen1Floods11NonGeoDataModule.

Parameters:

Name	Type	Description	Default
data_root	str	Root directory of the dataset.	required
batch_size	int	Batch size for DataLoaders. Defaults to 4.	4
num_workers	int	Number of workers for data loading. Defaults to 0.	0
bands	Sequence[str]	List of bands to use. Defaults to Sen1Floods11NonGeo.all_band_names.	all_band_names
train_transform	Compose | None | list[BasicTransform]	Transformations for training data.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation data.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for test data.	None
predict_transform	Compose | None | list[BasicTransform]	Transformations for prediction data.	None
drop_last	bool	Whether to drop the last incomplete batch. Defaults to True.	True
constant_scale	float	Scale constant applied to the dataset. Defaults to 0.0001.	0.0001
no_data_replace	float | None	Replacement value for missing data. Defaults to 0.	0
no_label_replace	int | None	Replacement value for missing labels. Defaults to -1.	-1
use_metadata	bool	Whether to return metadata info (time and location).	False
**kwargs	Any	Additional keyword arguments.	{}

setup(stage)

Set up datasets.

Parameters:

Name	Type	Description	Default
stage	str	Either fit, validate, test, or predict.	required

terratorch.datamodules.sen4agrinet

Sen4AgriNetDataModule

Bases: NonGeoDataModule

NonGeo LightningDataModule implementation for Sen4AgriNet.

__init__(bands=None, batch_size=8, num_workers=0, data_root='./', train_transform=None, val_transform=None, test_transform=None, predict_transform=None, seed=42, scenario='random', requires_norm=True, binary_labels=False, linear_encoder=None, **kwargs)

Initializes the Sen4AgriNetDataModule for the Sen4AgriNet dataset.

Parameters:

Name	Type	Description	Default
bands	list[str] | None	List of bands to use. Defaults to None.	None
batch_size	int	Batch size for DataLoaders. Defaults to 8.	8
num_workers	int	Number of workers for data loading. Defaults to 0.	0
data_root	str	Root directory of the dataset. Defaults to "./".	'./'
train_transform	Compose | None | list[BasicTransform]	Transformations for training data.	None
val_transform	Compose | None | list[BasicTransform]	Transformations for validation data.	None
test_transform	Compose | None | list[BasicTransform]	Transformations for test data.	None
predict_transform	Compose | None | list[BasicTransform]	Transformations for prediction data.	None
seed	int	Random seed for reproducibility. Defaults to 42.	42
scenario	str	Defines the splitting scenario to use. Options are: - 'random': Random split of the data. - 'spatial': Split by geographical regions (Catalonia and France). - 'spatio-temporal': Split by region and year (France 2019 and Catalonia 2020).	'random'
requires_norm	bool	Whether normalization is required. Defaults to True.	True
binary_labels	bool	Whether to use binary labels. Defaults to False.	False
linear_encoder	dict	Mapping for label encoding. Defaults to None.	None
**kwargs	Any	Additional keyword arguments.	{}

setup(stage)

Set up datasets.

Parameters:

Name	Type	Description	Default
stage	str	Either fit, validate, test, or predict.	required

terratorch.datamodules.sen4map

Sen4MapLucasDataModule

Bases: LightningDataModule

NonGeo LightningDataModule implementation for Sen4map.

__init__(batch_size, num_workers, prefetch_factor=0, train_hdf5_path=None, train_hdf5_keys_path=None, test_hdf5_path=None, test_hdf5_keys_path=None, val_hdf5_path=None, val_hdf5_keys_path=None, **kwargs)

Initializes the Sen4MapLucasDataModule for handling Sen4Map monthly composites.

Parameters:

Name	Type	Description	Default
batch_size	int	Batch size for DataLoaders.	required
num_workers	int	Number of worker processes for data loading.	required
prefetch_factor	int	Number of samples to prefetch per worker. Defaults to 0.	0
train_hdf5_path	str	Path to the training HDF5 file.	None
train_hdf5_keys_path	str	Path to the training HDF5 keys file.	None
test_hdf5_path	str	Path to the testing HDF5 file.	None
test_hdf5_keys_path	str	Path to the testing HDF5 keys file.	None
val_hdf5_path	str	Path to the validation HDF5 file.	None
val_hdf5_keys_path	str	Path to the validation HDF5 keys file.	None
train_hdf5_keys_save_path	str	(from kwargs) Path to save generated train keys.	required
test_hdf5_keys_save_path	str	(from kwargs) Path to save generated test keys.	required
val_hdf5_keys_save_path	str	(from kwargs) Path to save generated validation keys.	required
shuffle	bool	Global shuffle flag.	required
train_shuffle	bool	Shuffle flag for training data; defaults to global shuffle if unset.	required
val_shuffle	bool	Shuffle flag for validation data.	required
test_shuffle	bool	Shuffle flag for test data.	required
train_data_fraction	float	Fraction of training data to use. Defaults to 1.0.	required
val_data_fraction	float	Fraction of validation data to use. Defaults to 1.0.	required
test_data_fraction	float	Fraction of test data to use. Defaults to 1.0.	required
all_hdf5_data_path	str	General HDF5 data path for all splits. If provided, overrides specific paths.	required
resize	bool	Whether to resize images. Defaults to False.	required
resize_to	int or tuple	Target size for resizing images.	required
resize_interpolation	str	Interpolation mode for resizing ('bilinear', 'bicubic', etc.).	required
resize_antialiasing	bool	Whether to apply antialiasing during resizing. Defaults to True.	required
**kwargs		Additional keyword arguments.	{}

setup(stage)

Set up datasets.

Parameters:

Name	Type	Description	Default
stage	str	Either fit, test.	required

terratorch.datamodules.torchgeo_data_module

Ugly proxy objects so parsing config file works with transforms.

These are necessary since, for LightningCLI to instantiate arguments as objects from the config, they must have type annotations

In TorchGeo, transforms is passed in **kwargs, so it has no type annotations! To get around that, we create these wrappers that have transforms type annotated. They create the transforms and forward all method and attribute calls to the original TorchGeo datamodule.

Additionally, TorchGeo datasets pass the data to the transforms callable as a dict, and as a tensor.

Albumentations expects this data not as a dict but as different key-value arguments, and as numpy. We handle that conversion here.

TorchGeoDataModule

Bases: GeoDataModule

Proxy object for using Geo data modules defined by TorchGeo.

Allows for transforms to be defined and passed using config files. The only reason this class exists is so that we can annotate the transforms argument with a type. This is required for lightningcli and config files. As such, all getattr and setattr will be redirected to the underlying class.

__init__(cls, batch_size=None, num_workers=0, transforms=None, **kwargs)

Constructor

Parameters:

Name	Type	Description	Default
cls	type[GeoDataModule]	TorchGeo DataModule class to be instantiated	required
batch_size	int | None	batch_size. Defaults to None.	None
num_workers	int	num_workers. Defaults to 0.	0
transforms	None | list[BasicTransform]	List of Albumentations Transforms. Should enc with ToTensorV2. Defaults to None.	None
**kwargs	Any	Arguments passed to instantiate cls.	{}

TorchNonGeoDataModule

Bases: NonGeoDataModule

Proxy object for using NonGeo data modules defined by TorchGeo.

Allows for transforms to be defined and passed using config files. The only reason this class exists is so that we can annotate the transforms argument with a type. This is required for lightningcli and config files. As such, all getattr and setattr will be redirected to the underlying class.

__init__(cls, batch_size=None, num_workers=0, transforms=None, **kwargs)

Constructor

Parameters:

Name	Type	Description	Default
cls	type[NonGeoDataModule]	TorchGeo DataModule class to be instantiated	required
batch_size	int | None	batch_size. Defaults to None.	None
num_workers	int	num_workers. Defaults to 0.	0
transforms	None | list[BasicTransform]	List of Albumentations Transforms. Should enc with ToTensorV2. Defaults to None.	None
**kwargs	Any	Arguments passed to instantiate cls.	{}

---

Last update: March 23, 2025