Sursentia - aviary-models

View source

`Sursentia`

Tiles processor that uses the Sursentia model to predict landcover and solar panels.

Model input channels

ChannelName.R: Red channel, raster channel, ground sampling distance of 0.1 to 0.5 meters per pixel, standardized values with a mean of 0.392 and a standard deviation of 0.198
ChannelName.G: Green channel, raster channel, ground sampling distance of 0.1 to 0.5 meters per pixel, standardized values with a mean of 0.423 and a standard deviation of 0.173
ChannelName.B: Blue channel, raster channel, ground sampling distance of 0.1 to 0.5 meters per pixel, standardized values with a mean of 0.373 and a standard deviation of 0.157
Use the SursentiaPreprocessor to preprocess the input channels

Model output channels

'sursentia_landcover': Landcover channel, raster channel, ground sampling distance of the input channels, the values are 0 (buildings), 1 (buildings with green roofs), 2 (impervious surfaces), 3 (non-impervious surfaces), and 4 (water bodies)
'sursentia_solar': Solar panels channel, raster channel, ground sampling distance of the input channels, the values are 0 (background) and 1 (solar panels)
Use the SursentiaMapFieldProcessor to map the fields of the layers

Additional dependencies

Sursentia requires the sursentia dependency group and torch.

Implements the TilesProcessor protocol.

PARAMETER	DESCRIPTION
`r_channel_name`	Channel name of the red channel TYPE: `ChannelName \| str` DEFAULT: `ChannelName.R`
`g_channel_name`	Channel name of the green channel TYPE: `ChannelName \| str` DEFAULT: `ChannelName.G`
`b_channel_name`	Channel name of the blue channel TYPE: `ChannelName \| str` DEFAULT: `ChannelName.B`
`landcover_channel_name`	Channel name of the landcover channel (if None, the landcover head of the model is not used) TYPE: `str \| None` DEFAULT: `'sursentia_landcover'`
`solar_channel_name`	Channel name of the solar panels channel (if None, the solar head of the model is not used) TYPE: `str \| None` DEFAULT: `'sursentia_solar'`
`batch_size`	Batch size of the sliding window inference TYPE: `int` DEFAULT: `1`
`version`	Version of the model (`V1_0`) TYPE: `SursentiaVersion` DEFAULT: `SursentiaVersion.V1_0`
`device`	Device to run the model on (`CPU` or `GPU`) TYPE: `Device` DEFAULT: `Device.CPU`
`cache_dir_path`	Path to cache directory of the model TYPE: `Path` DEFAULT: `Path('cache')`
`remove_channels`	If True, the channels are removed TYPE: `bool` DEFAULT: `True`

`from_config` `classmethod`

Creates the Sursentia model from the configuration.

PARAMETER	DESCRIPTION
`config`	Configuration TYPE: `SursentiaConfig`

RETURNS	DESCRIPTION
`Sursentia`	Sursentia

`call`

Runs the Sursentia model.

PARAMETER	DESCRIPTION
`tiles`	Tiles TYPE: `Tiles`

RETURNS	DESCRIPTION
`Tiles`	Tiles

`SursentiaConfig`

Bases: pydantic.BaseModel

Configuration for the from_config class method of Sursentia

Create the configuration from a config file

Use '1.0' instead of SursentiaVersion.V1_0
Use 'cpu' or 'gpu' instead of Device.CPU or Device.GPU
Use null instead of None
Use false or true instead of False or True

Example

You can create the configuration from a config file.

config.yaml

package: 'aviary_models'
name: 'Sursentia'
config:
  r_channel_name: 'r'
  g_channel_name: 'g'
  b_channel_name: 'b'
  landcover_channel_name: 'sursentia_landcover'
  solar_channel_name: 'sursentia_solar'
  batch_size: 1
  version: '1.0'
  device: 'cpu'
  cache_dir_path: 'cache'
  remove_channels: true

ATTRIBUTE	DESCRIPTION
`r_channel_name`	Channel name of the red channel - defaults to `ChannelName.R` TYPE: `ChannelName \| str`
`g_channel_name`	Channel name of the green channel - defaults to `ChannelName.G` TYPE: `ChannelName \| str`
`b_channel_name`	Channel name of the blue channel - defaults to `ChannelName.B` TYPE: `ChannelName \| str`
`landcover_channel_name`	Channel name of the landcover channel (if None, the landcover head of the model is not used) - defaults to 'sursentia_landcover' TYPE: `str \| None`
`solar_channel_name`	Channel name of the solar panels channel (if None, the solar head of the model is not used) - defaults to 'sursentia_solar' TYPE: `str \| None`
`batch_size`	Batch size of the sliding window inference - defaults to 1 TYPE: `int`
`version`	Version of the model (`V1_0`) - defaults to `SursentiaVersion.V1_0` TYPE: `SursentiaVersion`
`device`	Device to run the model on (`CPU` or `GPU`) - defaults to `Device.CPU` TYPE: `Device`
`cache_dir_path`	Path to the cache directory of the model - defaults to 'cache' TYPE: `Path`
`remove_channels`	If True, the channels are removed - defaults to True TYPE: `bool`

`SursentiaPreprocessor`

Tiles processor that preprocesses the input channels of the Sursentia model.

Implements the TilesProcessor protocol.

PARAMETER	DESCRIPTION
`r_channel_name`	Channel name of the red channel TYPE: `ChannelName \| str` DEFAULT: `ChannelName.R`
`g_channel_name`	Channel name of the green channel TYPE: `ChannelName \| str` DEFAULT: `ChannelName.G`
`b_channel_name`	Channel name of the blue channel TYPE: `ChannelName \| str` DEFAULT: `ChannelName.B`
`new_r_channel_name`	New channel name of the red channel TYPE: `ChannelName \| str \| None` DEFAULT: `None`
`new_g_channel_name`	New channel name of the green channel TYPE: `ChannelName \| str \| None` DEFAULT: `None`
`new_b_channel_name`	New channel name of the blue channel TYPE: `ChannelName \| str \| None` DEFAULT: `None`
`max_num_threads`	Maximum number of threads TYPE: `int \| None` DEFAULT: `None`

`from_config` `classmethod`

Creates a sursentia preprocessor from the configuration.

PARAMETER	DESCRIPTION
`config`	Configuration TYPE: `SursentiaPreprocessorConfig`

RETURNS	DESCRIPTION
`SursentiaPreprocessor`	Sursentia preprocessor

`call`

Preprocesses the input channels.

PARAMETER	DESCRIPTION
`tiles`	Tiles TYPE: `Tiles`

RETURNS	DESCRIPTION
`Tiles`	Tiles

`SursentiaPreprocessorConfig`

Bases: pydantic.BaseModel

Configuration for the from_config class method of SursentiaPreprocessor

Create the configuration from a config file

Use null instead of None

Example

You can create the configuration from a config file.

config.yaml

package: 'aviary_models'
name: 'SursentiaPreprocessor'
config:
  r_channel_name: 'r'
  g_channel_name: 'g'
  b_channel_name: 'b'
  new_r_channel_name: null
  new_g_channel_name: null
  new_b_channel_name: null
  max_num_threads: null

ATTRIBUTE	DESCRIPTION
`r_channel_name`	Channel name of the red channel - defaults to `ChannelName.R` TYPE: `ChannelName \| str`
`g_channel_name`	Channel name of the green channel - defaults to `ChannelName.G` TYPE: `ChannelName \| str`
`b_channel_name`	Channel name of the blue channel - defaults to `ChannelName.B` TYPE: `ChannelName \| str`
`new_r_channel_name`	New channel name of the red channel - defaults to None TYPE: `ChannelName \| str \| None`
`new_g_channel_name`	New channel name of the green channel - defaults to None TYPE: `ChannelName \| str \| None`
`new_b_channel_name`	New channel name of the blue channel - defaults to None TYPE: `ChannelName \| str \| None`
`max_num_threads`	Maximum number of threads - defaults to None TYPE: `int \| None`

`SursentiaMapFieldProcessor`

Vector processor that maps the fields of the layers of the Sursentia model.

Landcover mapping

0: 'Gebäude'
1: 'Gründach'
2: 'versiegelte Fläche'
3: 'nicht versiegelte Fläche'
4: 'Gewässer'

Solar mapping

0: 'Hintergrund'
1: 'Solaranlage'

Implements the VectorProcessor protocol.

PARAMETER	DESCRIPTION
`field`	Field TYPE: `str`
`landcover_layer_name`	Layer name of the landcover layer (if None, the landcover layer is not used) TYPE: `str \| None` DEFAULT: `'sursentia_landcover'`
`solar_layer_name`	Layer name of the solar layer (if None, the solar layer is not used) TYPE: `str \| None` DEFAULT: `'sursentia_solar'`
`new_landcover_layer_name`	New layer name of the landcover layer TYPE: `str \| None` DEFAULT: `None`
`new_solar_layer_name`	New layer name of the solar layer TYPE: `str \| None` DEFAULT: `None`

`from_config` `classmethod`

Creates a sursentia map field processor from the configuration.

PARAMETER	DESCRIPTION
`config`	Configuration TYPE: `SursentiaMapFieldProcessorConfig`

RETURNS	DESCRIPTION
`SursentiaMapFieldProcessor`	Sursentia map field processor

`call`

Maps the fields of the layers.

PARAMETER	DESCRIPTION
`vector`	Vector TYPE: `Vector`

RETURNS	DESCRIPTION
`Vector`	Vector

`SursentiaMapFieldProcessorConfig`

Bases: pydantic.BaseModel

Configuration for the from_config class method of SursentiaMapFieldProcessor

Create the configuration from a config file

Use null instead of None

Example

You can create the configuration from a config file.

config.yaml

package: 'aviary_models'
name: 'SursentiaMapFieldProcessor'
config:
  field: 'my_field'
  landcover_layer_name: 'sursentia_landcover'
  solar_layer_name: 'sursentia_solar'
  new_landcover_layer_name: null
  new_solar_layer_name: null

ATTRIBUTE	DESCRIPTION
`field`	Field TYPE: `str`
`landcover_layer_name`	Layer name of the landcover layer (if None, the landcover layer is not used) - defaults to 'sursentia_landcover' TYPE: `str \| None`
`solar_layer_name`	Layer name of the solar layer (if None, the solar layer is not used) - defaults to 'sursentia_solar' TYPE: `str \| None`
`new_landcover_layer_name`	New layer name of the landcover layer - defaults to None TYPE: `str \| None`
`new_solar_layer_name`	New layer name of the solar layer - defaults to None TYPE: `str \| None`

`Device`

Bases: Enum

ATTRIBUTE	DESCRIPTION
`CPU`	CPU device
`GPU`	GPU device

`SursentiaVersion`

Bases: Enum

ATTRIBUTE	DESCRIPTION
`V1_0`	Version 1.0