vhr-cloudmask package

vhr_cloudmask.model.config.cloudmask_config

Classes:

CloudMaskConfig(data_dir, model_dir, model)

class vhr_cloudmask.model.config.cloudmask_config.CloudMaskConfig(data_dir, model_dir, model, inference_save_dir='results', experiment_name='unet-cnn', experiment_type='landcover', seed=24, gpu_devices='0, 1, 2, 3', mixed_precision=True, xla=False, input_bands=<factory>, output_bands=<factory>, modify_labels=None, substract_labels=False, expand_dims=True, tile_size=256, include_classes=False, augment=True, center_crop=False, no_data=0, nodata_fractional=False, nodata_fractional_tolerance=0.75, json_tiles_dir=None, dataset_from_json=False, normalize=1.0, normalize_label=1.0, rescale=None, standardization=None, batch_size=32, n_classes=1, test_size=0.2, mean=<factory>, std=<factory>, loss='tf.keras.losses.CategoricalCrossentropy', optimizer='tf.keras.optimizers.Adam', metrics=<factory>, callbacks=<factory>, transfer_learning=None, transfer_learning_weights=None, transfer_learning_fine_tune_at=None, learning_rate=0.0001, max_epochs=6000, patience=7, model_filename=None, inference_regex='*.tif', inference_regex_list=<factory>, window_size=8120, inference_overlap=0.5, inference_treshold=0.5, inference_pad_value=1000, window_algorithm='triang', pred_batch_size=128, probability_map=False, prediction_dtype='uint8', prediction_nodata=255, prediction_compress='LZW', prediction_driver='GTiff', metadata_regex=None, validation_database=None, test_classes=<factory>, test_colors=<factory>, test_truth_regex=None, hf_repo_id='nasa-cisto-data-science-group/vhr-cloudmask', hf_model_filename='cloudmask-vietnam-senegal-46-0.04.hdf5')

Bases: Config

Attributes:

hf_model_filename
hf_repo_id
test_classes

callbacks: List[str]

data_dir: str

hf_model_filename: str = 'cloudmask-vietnam-senegal-46-0.04.hdf5'

hf_repo_id: str = 'nasa-cisto-data-science-group/vhr-cloudmask'

inference_regex_list: Optional[List[str]]

input_bands: List[str]

mean: List[float]

metrics: List[str]

model: str

model_dir: Optional[str]

output_bands: List[str]

std: List[float]

test_classes: List[str]

test_colors: Optional[List[str]]

vhr_cloudmask.model.pipelines.cloudmask_cnn_pipeline

Classes:

CloudMaskPipeline([config_filename, ...])

This is a conceptual class representation of a CNN Segmentation TensorFlow pipeline.

class vhr_cloudmask.model.pipelines.cloudmask_cnn_pipeline.CloudMaskPipeline(config_filename=None, data_csv=None, model_filename=None, output_dir=None, inference_regex_list=None, default_config='templates/cloudmask_default.yaml', logger=None)

Bases: CNNSegmentation

This is a conceptual class representation of a CNN Segmentation TensorFlow pipeline. It is essentially an extended combination of the tensorflow_caney.model.pipelines.cnn_segmentation.CNNSegmentation.

Parameters:

• logger (str) – A logger device

• conf (omegaconf.OmegeConf object) – Configuration device

• data_csv (str) – CSV filename with data files for training

• experiment_name (str) – Experiment name description

• images_dir (str) – Directory to store training images

• labels_dir (str) – Directory to store training labels

• model_dir (str) – Directory to store trained models

Methods:

predict()

This will perform inference on a list of GeoTIFF files provided as a list of regexes from the CLI.

predict()

This will perform inference on a list of GeoTIFF files provided as a list of regexes from the CLI.

Returns:

None, outputs GeoTIFF cloudmask files to disk.

Return type:

None

vhr_cloudmask.view.cloudmask_cnn_pipeline_cli

Functions:

main()

vhr_cloudmask.view.cloudmask_cnn_pipeline_cli.main()