Segmentation IOProcessor Plugin

This plugin targets segmentation tasks and allows for the input image to be split in tiles of a size that depends on the model and on an arbitrary number of bands.

During initialization, the plugin accesses the model's data module configuration from the vLLM configuration and instantiates a DataModule object dynamically.

This plugin is installed as terratorch_segmentation.

Plugin specification

Plugin configuration

This plugin allows for additional configuration data to be passed via the TERRATORCH_SEGMENTATION_IO_PROCESSOR_CONFIG environment variable. If set, the variable should contain the plugin configuration in json string format.

The plugin configuration format is defined in the PluginConfig class.

terratorch.vllm.plugins.segmentation.types.PluginConfig

Bases: BaseModel

Source code in terratorch/vllm/plugins/segmentation/types.py

class PluginConfig(BaseModel):
    output_path: str = None
    """
    Default output folder path to be used when the out_data_format is set to path. 
    If omitted, the plugin will default to the current user home directory.
    """

    @model_validator(mode="after")
    def validate_values(self) -> Self:
        if not self.output_path:
            self.output_path = str(Path.home())
        elif os.path.exists(self.output_path):
            if not os.access(self.output_path, os.W_OK):
                raise ValueError(f"The path: {self.output_path} is not writable")
        else:
            raise ValueError(f"The path: {self.output_path} does not exist")

        return self

output_path = None class-attribute instance-attribute

Default output folder path to be used when the out_data_format is set to path. If omitted, the plugin will default to the current user home directory.

Request Data Format

The input format for for the plugin is defined in the RequestData class.

terratorch.vllm.plugins.segmentation.types.RequestData

Bases: BaseModel

Source code in terratorch/vllm/plugins/segmentation/types.py

class RequestData(BaseModel):

    data_format: Literal["b64_json", "path", "url"]
    """
    Data type for the input image.
    Allowed values are: [`b64_json`, `path`, `url`]
    """

    out_data_format: Literal["b64_json", "path"]
    """
    Data type for the output image.
    Allowed values are: [`b64_json`, `path`]
    """

    data: Any
    """
    Input image data
    """

    indices: Optional[list[int]] = None
    """
    Indices for bands to be processed in the input file
    """

data instance-attribute

Input image data

data_format instance-attribute

Data type for the input image. Allowed values are: [b64_json, path, url]

indices = None class-attribute instance-attribute

Indices for bands to be processed in the input file

out_data_format instance-attribute

Data type for the output image. Allowed values are: [b64_json, path]

Depending on the values set in data_format, the plugin expects data to contain a string that complies to the format. Similarly, out_data_format controls the data format returned to the user.

Request Output Format

The output format for the plugin is defined in the RequestOutput class.

terratorch.vllm.plugins.segmentation.types.RequestOutput

Bases: BaseModel

Source code in terratorch/vllm/plugins/segmentation/types.py

class RequestOutput(BaseModel):

    data_format: Literal["b64_json", "path"]
    """
    Data type for the output image.
    Allowed values are: [`b64_json`, `path`]
    """

    data: Any
    """
    Output image data
    """

    request_id: Optional[str] = None
    """
    The vLLM request ID if applicable
    """

data instance-attribute

Output image data

data_format instance-attribute

Data type for the output image. Allowed values are: [b64_json, path]

request_id = None class-attribute instance-attribute

The vLLM request ID if applicable