SemanticSegmentationLabelStore#

class SemanticSegmentationLabelStore[source]#

Bases: LabelStore

Storage for semantic segmentation predictions.

Can store predicted class ID raster and class scores raster as GeoTIFFs, and can optionally vectorize predictions and store them as GeoJSON files.

Attributes

`bbox`	Bounding box applied to the source.
`crs_transformer`	Associated `CRSTransformer`.
`hits_uri`
`label_uri`
`root_uri_local`
`score_uri`
`vector_output_dir_local`
`vector_output_uri`

__init__(uri: str, crs_transformer: CRSTransformer, class_config: ClassConfig, bbox: rastervision.core.box.Box | None = None, tmp_dir: str | None = None, vector_outputs: Sequence[VectorOutputConfig] | None = None, save_as_rgb: bool = False, discrete_output: bool = True, smooth_output: bool = False, smooth_as_uint8: bool = False, rasterio_block_size: int = 512)[source]#

Constructor.

Parameters

uri (str) – Path to directory where the predictions are/will be stored. Smooth scores will be saved as "uri/scores.tif", discrete labels will be stored as "uri/labels.tif", and vector outputs will be saved in "uri/vector_outputs/".
crs_transformer (CRSTransformer) – CRS transformer for correctly mapping from pixel coords to map coords.
class_config (ClassConfig) – Class config.
bbox (rastervision.core.box.Box | None) – User-specified crop of the extent. If provided, only labels falling inside it are returned by SemanticSegmentationLabelStore.get_labels(). Must be provided if the corresponding RasterSource has bbox != extent.
tmp_dir (str | None) – Temporary directory to use. If None, will be auto-generated. Defaults to None.
vector_outputs (Sequence[VectorOutputConfig] | None) – List of VectorOutputConfig’s containing vectorization configuration information. Only classes for which a VectorOutputConfig is specified will be saved as vectors. If None, no vector outputs will be produced. Defaults to None.
save_as_rgb (bool) – If True, saves labels as an RGB image, using the class-color mapping in the class_config. Defaults to False.
discrete_output (bool) – If True, saves labels as a raster of class IDs (one band). Defaults to False.
smooth_output (bool) – If True, saves labels as a raster of class scores (one band for each class). Defaults to False.
smooth_as_uint8 (bool) – If True, stores smooth class scores as np.uint8 (0-255) values rather than as np.float32 discrete labels, to help save memory/disk space. Defaults to False.
rasterio_block_size (int) – Value to set blockxsize and blockysize to. Defaults to 512.

Methods

`__init__`(uri, crs_transformer, class_config)	Constructor.
`build_rasterio_gtiff_profile`(labels)
`get_discrete_labels`()	Get all labels.
`get_labels`()	Get all labels.
`get_scores`()	Get all scores.
`merge_with_old_scores`(labels)
`save`(labels[, profile])	Save labels to disk.
`set_bbox`(bbox)	Set self.extent to the given value.
`write_discrete_raster_output`(out_profile, ...)
`write_smooth_raster_output`(out_profile, ...)
`write_vector_output`(vo, mask, vector_output_dir)	Write vector output for a single `VectorOutputConfig`.
`write_vector_outputs`(labels, vector_output_dir)	Write vectorized outputs for all configs in self.vector_outputs.

Constructor.

Parameters

uri (str) – Path to directory where the predictions are/will be stored. Smooth scores will be saved as "uri/scores.tif", discrete labels will be stored as "uri/labels.tif", and vector outputs will be saved in "uri/vector_outputs/".
crs_transformer (CRSTransformer) – CRS transformer for correctly mapping from pixel coords to map coords.
class_config (ClassConfig) – Class config.
bbox (rastervision.core.box.Box | None) – User-specified crop of the extent. If provided, only labels falling inside it are returned by SemanticSegmentationLabelStore.get_labels(). Must be provided if the corresponding RasterSource has bbox != extent.
tmp_dir (str | None) – Temporary directory to use. If None, will be auto-generated. Defaults to None.
vector_outputs (Sequence[VectorOutputConfig] | None) – List of VectorOutputConfig’s containing vectorization configuration information. Only classes for which a VectorOutputConfig is specified will be saved as vectors. If None, no vector outputs will be produced. Defaults to None.
save_as_rgb (bool) – If True, saves labels as an RGB image, using the class-color mapping in the class_config. Defaults to False.
discrete_output (bool) – If True, saves labels as a raster of class IDs (one band). Defaults to False.
smooth_output (bool) – If True, saves labels as a raster of class scores (one band for each class). Defaults to False.
smooth_as_uint8 (bool) – If True, stores smooth class scores as np.uint8 (0-255) values rather than as np.float32 discrete labels, to help save memory/disk space. Defaults to False.
rasterio_block_size (int) – Value to set blockxsize and blockysize to. Defaults to 512.

build_rasterio_gtiff_profile(labels: SemanticSegmentationLabels) → dict[source]#

Parameters: labels (SemanticSegmentationLabels) –
Return type: dict

get_discrete_labels() → SemanticSegmentationDiscreteLabels[source]#

Get all labels.

Returns: SemanticSegmentationDiscreteLabels
Return type: SemanticSegmentationDiscreteLabels

get_labels() → SemanticSegmentationLabels[source]#

Get all labels.

Returns: SemanticSegmentationLabels
Return type: SemanticSegmentationLabels

get_scores() → SemanticSegmentationSmoothLabels[source]#

Get all scores.

Returns: SemanticSegmentationSmoothLabels
Return type: SemanticSegmentationSmoothLabels

merge_with_old_scores(labels: SemanticSegmentationSmoothLabels) → SemanticSegmentationSmoothLabels[source]#

Parameters: labels (SemanticSegmentationSmoothLabels) –
Return type: SemanticSegmentationSmoothLabels

save(labels: SemanticSegmentationLabels, profile: dict | None = None) → None[source]#

Save labels to disk.

More info on rasterio IO: - https://github.com/mapbox/rasterio/blob/master/docs/quickstart.rst - https://rasterio.readthedocs.io/en/latest/topics/windowed-rw.html

Parameters

- (labels) –
labels (SemanticSegmentationLabels) –
profile (dict | None) –

Return type

None

set_bbox(bbox: Box) → None[source]#

Set self.extent to the given value.

Note

This method is idempotent.

Parameters

extent (Box) – User-specified extent in pixel coordinates.
bbox (Box) –

Return type

None

write_discrete_raster_output(out_profile: dict, path: str, labels: SemanticSegmentationLabels) → None[source]#

Parameters

out_profile (dict) –
path (str) –
labels (SemanticSegmentationLabels) –

Return type

None

write_smooth_raster_output(out_profile: dict, scores_path: str, hits_path: str, labels: SemanticSegmentationSmoothLabels) → None[source]#

Parameters

out_profile (dict) –
scores_path (str) –
hits_path (str) –
labels (SemanticSegmentationSmoothLabels) –

Return type

None

write_vector_output(vo: VectorOutputConfig, mask: ndarray, vector_output_dir: str) → None[source]#

Write vector output for a single VectorOutputConfig.

Parameters

vo (VectorOutputConfig) –
mask (ndarray) –
vector_output_dir (str) –

Return type

None

write_vector_outputs(labels: SemanticSegmentationLabels, vector_output_dir: str) → None[source]#

Write vectorized outputs for all configs in self.vector_outputs.

Parameters

labels (SemanticSegmentationLabels) –
vector_output_dir (str) –

Return type

None

property bbox: Box#: Bounding box applied to the source.

property crs_transformer: CRSTransformer#: Associated CRSTransformer.

property hits_uri: str#

property label_uri: str#

property root_uri_local: str#

property score_uri: str#

property vector_output_dir_local: str#

property vector_output_uri: str#