Skip to main content
Public Preview

DatasetEnum

Built-in imagery configurations available for RasterFlow mosaic creation.

Values

str
National Agriculture Imagery Program aerial imagery at 30 cm resolution, with red, green, blue, and near-infrared bands.
str
National Agriculture Imagery Program aerial imagery at 60 cm resolution, with red, green, blue, and near-infrared bands.
str
Three-band NAIP visualization COGs at 30 cm resolution, with red, green, and blue bands. The source uses lossy JPEG compression.
str
Three-band NAIP visualization COGs at 60 cm resolution, with red, green, and blue bands. The source uses lossy JPEG compression.
str
One Sentinel-2 L2A composite labelled by the required target_date within the inclusive [start, end] candidate window. Selects the nearest SCL-valid observation per pixel among selected scenes; earlier observations win ties. Provides B02, B03, B04, and B08 at 10 m. Source dates may differ per pixel.
str
Two closest-valid composites labelled by start and end, both using the inclusive [start, end] candidate window. Uses SCL and date proximity, without atmospheric-quality ranking. Provides B02, B03, B04, and B08 at 10 m.
str
Sentinel-2 L2A per-pixel median over UTC [start, end), labelled by start. Provides B02, B03, B04, and B08 at 10 m. Annual grouping is optional.
str
Same median and time rules as S2_MEDIAN, with all 12 L2A reflectance bands resampled to 10 m.
str
Sentinel-2 L2A median composite for a latitude-based harvest-season window. Provides the B02, B03, B04, and B08 bands at 10 m resolution, plus an N_VALID_PIXELS band.
str
Sentinel-2 L2A median composite for a latitude-based planting-season window. Provides the B02, B03, B04, and B08 bands at 10 m resolution, plus an N_VALID_PIXELS band.

DistanceMetricEnum

Distance metrics for comparing embedding vectors across time.

Values

str
Dequantize and normalize AlphaEarth int8 embeddings before computing cosine distance.
str
Compute one minus the cosine similarity of normalized embedding vectors.
str
Compute straight-line distance between embedding vectors.

GeometryActorEnum

Model task types that produce georeferenced geometries from raster inputs.

Values

str
Detect objects described by text prompts and return bounding-box geometries.
str
Detect objects described by text prompts and return polygon geometries.

GeometryModelRecipes

Preconfigured RasterFlow model recipes for text-prompted geometry outputs.

Values

str
Detect objects described by text prompts in 30 cm NAIP imagery and return georeferenced bounding boxes with confidence scores.
str
Detect objects described by text prompts in 30 cm NAIP imagery and return georeferenced polygon geometries with confidence scores.

GeometryOutput

Output reference for geometry inference workflow artifacts.

Attributes

str | None
URI to the output geometry parquet directory, or None if no geometries were detected. The parquet contains one row per detected geometry with the following columns:
  • geometry: georeferenced vector polygon
  • label: detected category name (e.g. the text prompt used)
  • bbox_score: detection confidence score (float)
  • bbox: bounding box dict with xmin, xmax, ymin, ymax keys
  • time: timestamp of the source imagery used for inference
  • source_store: URI of the source Zarr store the geometry was derived from
  • x_offset: tile x offset in the source store (pixels)
  • y_offset: tile y offset in the source store (pixels)

InferenceConfig

Configuration for model inference in RasterFlow.

Attributes

str
Path to the model file used for inference. Must be a Pytorch 2 Model Archive and have the extension .pt2. See https://docs.pytorch.org/docs/stable/export/pt2_archive.html for more information.
MosaicToMosaicActorEnum
The inference actor type to use for running the model.
int
Size of the patches to be used during inference in pixel units.
int
Size of the clip to be used during inference in pixel units.
str
Device to run the model on (e.g., “cuda”, “cpu”).
list[str] | list[int]
List of feature names or band indices to be used for the model inference. When using names, this must match the underlying band labels of the data source.
list[str]
List of label names to be used for the model inference.
int
Maximum batch size to use during inference.
MergeModeEnum
The mode to use for merging overlapping patches.

MergeModeEnum

Methods for assembling predictions from image patches.

Values

str
Remove the configured border from each patch before assembling predictions.
str
Assemble patch predictions without overlap-specific clipping or blending.
str
Blend overlapping patch predictions with spatial weights to reduce seams.

ModelRecipes

Preconfigured RasterFlow model recipes for raster prediction outputs.

Values

str
Detect rural roads in 1 m NAIP imagery with ChesapeakeRSC semantic segmentation.
str
Detect agricultural fields and field boundaries with Fields of the World, using Sentinel-2 planting- and harvest-season composites.
str
Estimate tree canopy height in meters from 60 cm NAIP RGB imagery with Meta CHM v1.
str
Detect sidewalks, roads, and crosswalks in 30 cm NAIP RGB imagery with Tile2Net semantic segmentation.

MosaicResult

Result for workflows that produce a mosaic index.

Attributes

any
URI of the mosaic index GeoParquet output.
any
URI of the first mosaic store referenced by the index, when available.
any
Sedona DataFrame read from mosaic_index_uri through the active Spark session.
any
Lazily loaded GeoDataFrame read from mosaic_index_uri. Requires GeoPandas.

MosaicToMosaicActorEnum

Model task types that produce raster prediction outputs from raster inputs.

Values

str
Produce continuous numeric predictions for each output pixel.
str
Stack multiple time steps as model inputs and produce one semantic-segmentation output for change detection.
str
Produce per-pixel class probabilities with a semantic-segmentation model.
str
Produce embedding bands from time-stamped imagery with a temporal embedding model.

ResamplingMethod

Methods for calculating output pixels when RasterFlow resizes imagery.

Values

str
Compute the weighted average of contributing pixels that contain data.
str
Interpolate from nearby source pixels with a bilinear kernel.
str
Interpolate from nearby source pixels with a cubic convolution kernel.
str
Interpolate from nearby source pixels with a cubic B-spline kernel.
str
Apply a Gaussian kernel when downsampling imagery.
str
Interpolate from nearby source pixels with a Lanczos windowed-sinc kernel.
str
Select the nearest source pixel. This is the fastest option and preserves categorical values.
str
Compute the root mean square of contributing pixels that contain data.

SemSegRasterioConfig

Configuration for vectorizing semantic segmentation rasters.

Attributes

bool
Whether to include score_mean and score_std confidence columns for each output feature. Defaults to False.
bool
Whether to thin detected regions toward their centerlines before creating geometries. Defaults to False.

TemporalScoreMethodEnum

Reference methods for scoring how each time step differs from a time series.

Values

str
Compare each time step with the immediately preceding time step.
str
Compare each time step with the mean of all other time steps.
str
Compare each time step with the medoid of all other time steps.

UriOutput

Base output reference carrying a URI to a generated artifact.

Attributes

str | None
URI to the produced artifact, or None when no output was produced.
str | None
URL to view this artifact in the Wherobots Cloud map viewer.

VectorizeMethodEnum

Methods for converting raster prediction bands to vector geometries.

Values

str
Threshold each semantic-segmentation score band, group connected foreground pixels, and convert those groups to polygon geometries.

VectorizeOutput

Output reference for vectorization workflows.

Attributes

str | None
URI to the merged parquet directory containing vectorization results. This output includes a source_store column and may be None when no vector features were produced.