from typing import TYPE_CHECKING, List
import logging
from rasterio.features import rasterize
import numpy as np
import geopandas as gpd
from rastervision.core.data.raster_source import RasterSource
log = logging.getLogger(__name__)
if TYPE_CHECKING:
from rastervision.core.box import Box
from rastervision.core.data import VectorSource, RasterTransformer
[docs]def geoms_to_raster(df: gpd.GeoDataFrame, window: 'Box',
background_class_id: int, all_touched: bool) -> np.ndarray:
"""Rasterize geometries that intersect with the window.
Args:
df (gpd.GeoDataFrame): All label geometries in the scene.
window (Box): The part of the scene to rasterize.
background_class_id (int): Class ID to use for pixels that don't
fall under any label geometry.
all_touched (bool): If True, all pixels touched by geometries will be
burned in. If false, only pixels whose center is within the
polygon or that are selected by Bresenham's line algorithm will be
burned in. (See :func:`.rasterize` for more details).
Defaults to False.
Returns:
np.ndarray: A raster.
"""
if len(df) == 0:
return np.full(window.size, background_class_id, dtype=np.uint8)
window_geom = window.to_shapely()
# subset to shapes that intersect window
df_int = df[df.intersects(window_geom)]
# transform to window frame of reference
shapes = df_int.translate(xoff=-window.xmin, yoff=-window.ymin)
# class IDs of each shape
class_ids = df_int['class_id']
if len(shapes) > 0:
raster = rasterize(
shapes=list(zip(shapes, class_ids)),
out_shape=window.size,
fill=background_class_id,
dtype=np.uint8,
all_touched=all_touched)
else:
raster = np.full(window.size, background_class_id, dtype=np.uint8)
return raster
[docs]class RasterizedSource(RasterSource):
"""A :class:`.RasterSource` based on the rasterization of a VectorSource."""
[docs] def __init__(self,
vector_source: 'VectorSource',
background_class_id: int,
extent: 'Box',
all_touched: bool = False,
raster_transformers: List['RasterTransformer'] = []):
"""Constructor.
Args:
vector_source (VectorSource): The VectorSource to rasterize.
background_class_id (int): The class_id to use for any background
pixels, ie. pixels not covered by a polygon.
extent (Box): Extent of corresponding imagery RasterSource.
all_touched (bool, optional): If True, all pixels touched by
geometries will be burned in. If false, only pixels whose
center is within the polygon or that are selected by
Bresenham's line algorithm will be burned in.
(See :func:`~rasterio.features.rasterize` for more details).
Defaults to False.
"""
self.vector_source = vector_source
self.background_class_id = background_class_id
self.all_touched = all_touched
self.df = self.vector_source.get_dataframe()
self.validate_labels(self.df)
super().__init__(
channel_order=[0],
num_channels_raw=1,
raster_transformers=raster_transformers,
extent=extent)
@property
def dtype(self) -> np.dtype:
return np.uint8
@property
def crs_transformer(self):
return self.vector_source.crs_transformer
def _get_chip(self, window: 'Box') -> np.ndarray:
"""Return the chip located in the window.
Polygons falling within the window are rasterized using their
``class_id`` property and the background is filled with
``background_class_id``.
Args:
window (Box): Window to read.
Returns:
np.ndarray: [height, width, channels] numpy array
"""
log.debug(f'Rasterizing window: {window}')
chip = geoms_to_raster(
self.df,
window,
background_class_id=self.background_class_id,
all_touched=self.all_touched)
# Add third singleton dim since rasters must have >=1 channel.
return np.expand_dims(chip, 2)
[docs] def validate_labels(self, df: gpd.GeoDataFrame) -> None:
"""Validate label geometries.
Args:
df (gpd.GeoDataFrame): Label geometries.
Raises:
ValueError: If ``Point`` or ``LineString`` geometries found.
ValueError: If geometries are missing class IDs.
"""
geom_types = set(df.geom_type)
if 'Point' in geom_types or 'LineString' in geom_types:
raise ValueError('LineStrings and Points are not supported '
'in RasterizedSource. Use BufferTransformer '
'to buffer them into Polygons. '
f'Geom types found in data: {geom_types}')
if len(df) > 0 and 'class_id' not in df.columns:
raise ValueError('All label polygons must have a class_id.')