Source code for torchgeo.datasets.enviroatlas
# Copyright (c) Microsoft Corporation. All rights reserved.
# Licensed under the MIT License.
"""EnviroAtlas High-Resolution Land Cover datasets."""
import os
import sys
from collections.abc import Callable, Sequence
from typing import Any, cast
import fiona
import matplotlib.pyplot as plt
import numpy as np
import pyproj
import rasterio
import rasterio.mask
import shapely.geometry
import shapely.ops
import torch
from matplotlib.colors import ListedColormap
from matplotlib.figure import Figure
from rasterio.crs import CRS
from .geo import GeoDataset
from .utils import BoundingBox, DatasetNotFoundError, download_url, extract_archive
[docs]class EnviroAtlas(GeoDataset):
"""EnviroAtlas dataset covering four cities with prior and weak input data layers.
The `EnviroAtlas
<https://doi.org/10.5281/zenodo.5778192>`__ dataset contains NAIP aerial imagery,
NLCD land cover labels, OpenStreetMap roads, water, waterways, and waterbodies,
Microsoft building footprint labels, high-resolution land cover labels from the
EPA EnviroAtlas dataset, and high-resolution land cover prior layers.
This dataset was organized to accompany the 2022 paper, `"Resolving label
uncertainty with implicit generative models"
<https://openreview.net/forum?id=AEa_UepnMDX>`_. More details can be found at
https://github.com/estherrolf/implicit-posterior.
If you use this dataset in your research, please cite the following paper:
* https://openreview.net/forum?id=AEa_UepnMDX
.. versionadded:: 0.3
"""
url = "https://zenodo.org/record/5778193/files/enviroatlas_lotp.zip?download=1"
filename = "enviroatlas_lotp.zip"
md5 = "bfe601be21c7c001315fc6154be8ef14"
crs = CRS.from_epsg(3857)
res = 1
valid_prior_layers = ["prior", "prior_no_osm_no_buildings"]
valid_layers = [
"naip",
"nlcd",
"roads",
"water",
"waterways",
"waterbodies",
"buildings",
"lc",
] + valid_prior_layers
cities = [
"pittsburgh_pa-2010_1m",
"durham_nc-2012_1m",
"austin_tx-2012_1m",
"phoenix_az-2010_1m",
]
splits = (
[f"{state}-train" for state in cities[:1]]
+ [f"{state}-val" for state in cities[:1]]
+ [f"{state}-test" for state in cities]
+ [f"{state}-val5" for state in cities]
)
# these are used to check the integrity of the dataset
_files = [
"austin_tx-2012_1m-test_tiles-debuffered",
"austin_tx-2012_1m-val5_tiles-debuffered",
"durham_nc-2012_1m-test_tiles-debuffered",
"durham_nc-2012_1m-val5_tiles-debuffered",
"phoenix_az-2010_1m-test_tiles-debuffered",
"phoenix_az-2010_1m-val5_tiles-debuffered",
"pittsburgh_pa-2010_1m-test_tiles-debuffered",
"pittsburgh_pa-2010_1m-train_tiles-debuffered",
"pittsburgh_pa-2010_1m-val5_tiles-debuffered",
"pittsburgh_pa-2010_1m-val_tiles-debuffered",
"austin_tx-2012_1m-test_tiles-debuffered/3009726_sw_a_naip.tif",
"austin_tx-2012_1m-test_tiles-debuffered/3009726_sw_b_nlcd.tif",
"austin_tx-2012_1m-test_tiles-debuffered/3009726_sw_c_roads.tif",
"austin_tx-2012_1m-test_tiles-debuffered/3009726_sw_d1_waterways.tif",
"austin_tx-2012_1m-test_tiles-debuffered/3009726_sw_d2_waterbodies.tif",
"austin_tx-2012_1m-test_tiles-debuffered/3009726_sw_d_water.tif",
"austin_tx-2012_1m-test_tiles-debuffered/3009726_sw_e_buildings.tif",
"austin_tx-2012_1m-test_tiles-debuffered/3009726_sw_h_highres_labels.tif",
"austin_tx-2012_1m-test_tiles-debuffered/3009726_sw_prior_from_cooccurrences_101_31.tif", # noqa: E501
"austin_tx-2012_1m-test_tiles-debuffered/3009726_sw_prior_from_cooccurrences_101_31_no_osm_no_buildings.tif", # noqa: E501
"spatial_index.geojson",
]
p_src_crs = pyproj.CRS("epsg:3857")
p_transformers = {
"epsg:26917": pyproj.Transformer.from_crs(
p_src_crs, pyproj.CRS("epsg:26917"), always_xy=True
).transform,
"epsg:26918": pyproj.Transformer.from_crs(
p_src_crs, pyproj.CRS("epsg:26918"), always_xy=True
).transform,
"epsg:26914": pyproj.Transformer.from_crs(
p_src_crs, pyproj.CRS("epsg:26914"), always_xy=True
).transform,
"epsg:26912": pyproj.Transformer.from_crs(
p_src_crs, pyproj.CRS("epsg:26912"), always_xy=True
).transform,
}
# used to convert the 10 high-res classes labeled as [0, 10, 20, 30, 40, 52, 70, 80,
# 82, 91, 92] to sequential labels [0, ..., 10]
raw_enviroatlas_to_idx_map: "np.typing.NDArray[np.uint8]" = np.array(
[
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
2,
0,
0,
0,
0,
0,
0,
0,
0,
0,
3,
0,
0,
0,
0,
0,
0,
0,
0,
0,
4,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
5,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
6,
0,
0,
0,
0,
0,
0,
0,
0,
0,
7,
0,
8,
0,
0,
0,
0,
0,
0,
0,
0,
9,
10,
],
dtype=np.uint8,
)
highres_classes = [
"Unclassified",
"Water",
"Impervious Surface",
"Soil and Barren",
"Trees and Forest",
"Shrubs",
"Grass and Herbaceous",
"Agriculture",
"Orchards",
"Woody Wetlands",
"Emergent Wetlands",
]
highres_cmap = ListedColormap(
[
[1.00000000, 1.00000000, 1.00000000],
[0.00000000, 0.77254902, 1.00000000],
[0.61176471, 0.61176471, 0.61176471],
[1.00000000, 0.66666667, 0.00000000],
[0.14901961, 0.45098039, 0.00000000],
[0.80000000, 0.72156863, 0.47450980],
[0.63921569, 1.00000000, 0.45098039],
[0.86274510, 0.85098039, 0.22352941],
[0.67058824, 0.42352941, 0.15686275],
[0.72156863, 0.85098039, 0.92156863],
[0.42352941, 0.62352941, 0.72156863],
]
)
[docs] def __init__(
self,
root: str = "data",
splits: Sequence[str] = ["pittsburgh_pa-2010_1m-train"],
layers: Sequence[str] = ["naip", "prior"],
transforms: Callable[[dict[str, Any]], dict[str, Any]] | None = None,
prior_as_input: bool = False,
cache: bool = True,
download: bool = False,
checksum: bool = False,
) -> None:
"""Initialize a new Dataset instance.
Args:
root: root directory where dataset can be found
splits: a list of strings in the format "{state}-{train,val,test}"
indicating the subset of data to use, for example "ny-train"
layers: a list containing a subset of ``valid_layers`` indicating which
layers to load
transforms: a function/transform that takes an input sample
and returns a transformed version
prior_as_input: bool describing whether the prior is used as an input (True)
or as supervision (False)
cache: if True, cache file handle to speed up repeated sampling
download: if True, download dataset and store it in the root directory
checksum: if True, check the MD5 of the downloaded files (may be slow)
Raises:
AssertionError: if ``splits`` or ``layers`` are not valid
DatasetNotFoundError: If dataset is not found and *download* is False.
"""
for split in splits:
assert split in self.splits
assert all([layer in self.valid_layers for layer in layers])
self.root = root
self.layers = layers
self.cache = cache
self.download = download
self.checksum = checksum
self.prior_as_input = prior_as_input
self._verify()
super().__init__(transforms)
# Add all tiles into the index in epsg:3857 based on the included geojson
mint: float = 0
maxt: float = sys.maxsize
with fiona.open(
os.path.join(root, "enviroatlas_lotp", "spatial_index.geojson"), "r"
) as f:
for i, row in enumerate(f):
if row["properties"]["split"] in splits:
box = shapely.geometry.shape(row["geometry"])
minx, miny, maxx, maxy = box.bounds
coords = (minx, maxx, miny, maxy, mint, maxt)
self.index.insert(
i,
coords,
{
"naip": row["properties"]["naip"],
"nlcd": row["properties"]["nlcd"],
"roads": row["properties"]["roads"],
"water": row["properties"]["water"],
"waterways": row["properties"]["waterways"],
"waterbodies": row["properties"]["waterbodies"],
"buildings": row["properties"]["buildings"],
"lc": row["properties"]["lc"],
"prior_no_osm_no_buildings": row["properties"][
"naip"
].replace(
"a_naip",
"prior_from_cooccurrences_101_31_no_osm_no_buildings",
),
"prior": row["properties"]["naip"].replace(
"a_naip", "prior_from_cooccurrences_101_31"
),
},
)
[docs] def __getitem__(self, query: BoundingBox) -> dict[str, Any]:
"""Retrieve image/mask and metadata indexed by query.
Args:
query: (minx, maxx, miny, maxy, mint, maxt) coordinates to index
Returns:
sample of image/mask and metadata at that index
Raises:
IndexError: if query is not found in the index
"""
hits = self.index.intersection(tuple(query), objects=True)
filepaths = cast(list[dict[str, str]], [hit.object for hit in hits])
sample = {"image": [], "mask": [], "crs": self.crs, "bbox": query}
if len(filepaths) == 0:
raise IndexError(
f"query: {query} not found in index with bounds: {self.bounds}"
)
elif len(filepaths) == 1:
filenames = filepaths[0]
query_geom_transformed = None # is set by the first layer
minx, maxx, miny, maxy, mint, maxt = query
query_box = shapely.geometry.box(minx, miny, maxx, maxy)
for layer in self.layers:
fn = filenames[layer]
with rasterio.open(
os.path.join(self.root, "enviroatlas_lotp", fn)
) as f:
dst_crs = f.crs.to_string().lower()
if query_geom_transformed is None:
query_box_transformed = shapely.ops.transform(
self.p_transformers[dst_crs], query_box
).envelope
query_geom_transformed = shapely.geometry.mapping(
query_box_transformed
)
data, _ = rasterio.mask.mask(
f, [query_geom_transformed], crop=True, all_touched=True
)
if layer in [
"naip",
"buildings",
"roads",
"waterways",
"waterbodies",
"water",
]:
sample["image"].append(data)
elif layer in ["prior", "prior_no_osm_no_buildings"]:
if self.prior_as_input:
sample["image"].append(data)
else:
sample["mask"].append(data)
elif layer in ["lc"]:
data = self.raw_enviroatlas_to_idx_map[data]
sample["mask"].append(data)
else:
raise IndexError(f"query: {query} spans multiple tiles which is not valid")
sample["image"] = np.concatenate(sample["image"], axis=0)
sample["mask"] = np.concatenate(sample["mask"], axis=0)
sample["image"] = torch.from_numpy(sample["image"])
sample["mask"] = torch.from_numpy(sample["mask"])
if self.transforms is not None:
sample = self.transforms(sample)
return sample
def _verify(self) -> None:
"""Verify the integrity of the dataset."""
def exists(filename: str) -> bool:
return os.path.exists(os.path.join(self.root, "enviroatlas_lotp", filename))
# Check if the extracted files already exist
if all(map(exists, self._files)):
return
# Check if the zip files have already been downloaded
if os.path.exists(os.path.join(self.root, self.filename)):
self._extract()
return
# Check if the user requested to download the dataset
if not self.download:
raise DatasetNotFoundError(self)
# Download the dataset
self._download()
self._extract()
def _download(self) -> None:
"""Download the dataset."""
download_url(self.url, self.root, filename=self.filename, md5=self.md5)
def _extract(self) -> None:
"""Extract the dataset."""
extract_archive(os.path.join(self.root, self.filename))
[docs] def plot(
self,
sample: dict[str, Any],
show_titles: bool = True,
suptitle: str | None = None,
) -> Figure:
"""Plot a sample from the dataset.
Note: only plots the "naip" and "lc" layers.
Args:
sample: a sample returned by :meth:`__getitem__`
show_titles: flag indicating whether to show titles above each panel
suptitle: optional string to use as a suptitle
Returns:
a matplotlib Figure with the rendered sample
Raises:
ValueError: if the NAIP layer isn't included in ``self.layers``
"""
if "naip" not in self.layers or "lc" not in self.layers:
raise ValueError("The 'naip' and 'lc' layers must be included for plotting")
image_layers = []
mask_layers = []
for layer in self.layers:
if layer in [
"naip",
"buildings",
"roads",
"waterways",
"waterbodies",
"water",
]:
image_layers.append(layer)
elif layer in ["prior", "prior_no_osm_no_buildings"]:
if self.prior_as_input:
image_layers.append(layer)
else:
mask_layers.append(layer)
elif layer in ["lc"]:
mask_layers.append(layer)
naip_index = image_layers.index("naip")
lc_index = mask_layers.index("lc")
image = np.rollaxis(
sample["image"][naip_index : naip_index + 3, :, :].numpy(), 0, 3
)
mask = sample["mask"][lc_index].numpy()
num_panels = 2
showing_predictions = "prediction" in sample
if showing_predictions:
predictions = sample["prediction"].numpy()
num_panels += 1
fig, axs = plt.subplots(1, num_panels, figsize=(num_panels * 4, 5))
axs[0].imshow(image)
axs[0].axis("off")
axs[1].imshow(
mask, vmin=0, vmax=10, cmap=self.highres_cmap, interpolation="none"
)
axs[1].axis("off")
if show_titles:
axs[0].set_title("Image")
axs[1].set_title("Mask")
if showing_predictions:
axs[2].imshow(
predictions,
vmin=0,
vmax=10,
cmap=self.highres_cmap,
interpolation="none",
)
axs[2].axis("off")
if show_titles:
axs[2].set_title("Predictions")
if suptitle is not None:
plt.suptitle(suptitle)
return fig
