Source code for torchgeo.datasets.dfc2022
# Copyright (c) Microsoft Corporation. All rights reserved.
# Licensed under the MIT License.
"""2022 IEEE GRSS Data Fusion Contest (DFC2022) dataset."""
import glob
import os
from collections.abc import Callable, Sequence
import matplotlib.pyplot as plt
import numpy as np
import rasterio
import torch
from matplotlib import colors
from matplotlib.figure import Figure
from rasterio.enums import Resampling
from torch import Tensor
from .geo import NonGeoDataset
from .utils import (
DatasetNotFoundError,
check_integrity,
extract_archive,
percentile_normalization,
)
[docs]class DFC2022(NonGeoDataset):
"""DFC2022 dataset.
The `DFC2022 <https://www.grss-ieee.org/community/technical-committees/2022-ieee-grss-data-fusion-contest/>`__
dataset is used as a benchmark dataset for the 2022 IEEE GRSS Data Fusion Contest
and extends the MiniFrance dataset for semi-supervised semantic segmentation.
The dataset consists of a train set containing labeled and unlabeled imagery and an
unlabeled validation set. The dataset can be downloaded from the
`IEEEDataPort DFC2022 website <https://ieee-dataport.org/competitions/data-fusion-contest-2022-dfc2022/>`_.
Dataset features:
* RGB aerial images at 0.5 m per pixel spatial resolution (~2,000x2,0000 px)
* DEMs at 1 m per pixel spatial resolution (~1,000x1,0000 px)
* Masks at 0.5 m per pixel spatial resolution (~2,000x2,0000 px)
* 16 land use/land cover categories
* Images collected from the
`IGN BD ORTHO database <https://geoservices.ign.fr/documentation/donnees/ortho/bdortho/>`_
* DEMs collected from the
`IGN RGE ALTI database <https://geoservices.ign.fr/documentation/donnees/alti/rgealti/>`_
* Labels collected from the
`UrbanAtlas 2012 database <https://land.copernicus.eu/local/urban-atlas/urban-atlas-2012/view/>`_
* Data collected from 19 regions in France
Dataset format:
* images are three-channel geotiffs
* DEMS are single-channel geotiffs
* masks are single-channel geotiffs with the pixel values represent the class
Dataset classes:
0. No information
1. Urban fabric
2. Industrial, commercial, public, military, private and transport units
3. Mine, dump and construction sites
4. Artificial non-agricultural vegetated areas
5. Arable land (annual crops)
6. Permanent crops
7. Pastures
8. Complex and mixed cultivation patterns
9. Orchards at the fringe of urban classes
10. Forests
11. Herbaceous vegetation associations
12. Open spaces with little or no vegetation
13. Wetlands
14. Water
15. Clouds and Shadows
If you use this dataset in your research, please cite the following paper:
* https://doi.org/10.1007/s10994-020-05943-y
.. versionadded:: 0.3
""" # noqa: E501
classes = [
"No information",
"Urban fabric",
"Industrial, commercial, public, military, private and transport units",
"Mine, dump and construction sites",
"Artificial non-agricultural vegetated areas",
"Arable land (annual crops)",
"Permanent crops",
"Pastures",
"Complex and mixed cultivation patterns",
"Orchards at the fringe of urban classes",
"Forests",
"Herbaceous vegetation associations",
"Open spaces with little or no vegetation",
"Wetlands",
"Water",
"Clouds and Shadows",
]
colormap = [
"#231F20",
"#DB5F57",
"#DB9757",
"#DBD057",
"#ADDB57",
"#75DB57",
"#7BC47B",
"#58B158",
"#D4F6D4",
"#B0E2B0",
"#008000",
"#58B0A7",
"#995D13",
"#579BDB",
"#0062FF",
"#231F20",
]
metadata = {
"train": {
"filename": "labeled_train.zip",
"md5": "2e87d6a218e466dd0566797d7298c7a9",
"directory": "labeled_train",
},
"train-unlabeled": {
"filename": "unlabeled_train.zip",
"md5": "1016d724bc494b8c50760ae56bb0585e",
"directory": "unlabeled_train",
},
"val": {
"filename": "val.zip",
"md5": "6ddd9c0f89d8e74b94ea352d4002073f",
"directory": "val",
},
}
image_root = "BDORTHO"
dem_root = "RGEALTI"
target_root = "UrbanAtlas"
[docs] def __init__(
self,
root: str = "data",
split: str = "train",
transforms: Callable[[dict[str, Tensor]], dict[str, Tensor]] | None = None,
checksum: bool = False,
) -> None:
"""Initialize a new DFC2022 dataset instance.
Args:
root: root directory where dataset can be found
split: one of "train" or "test"
transforms: a function/transform that takes input sample and its target as
entry and returns a transformed version
checksum: if True, check the MD5 of the downloaded files (may be slow)
Raises:
AssertionError: if ``split`` is invalid
DatasetNotFoundError: If dataset is not found.
"""
assert split in self.metadata
self.root = root
self.split = split
self.transforms = transforms
self.checksum = checksum
self._verify()
self.class2idx = {c: i for i, c in enumerate(self.classes)}
self.files = self._load_files()
[docs] def __getitem__(self, index: int) -> dict[str, Tensor]:
"""Return an index within the dataset.
Args:
index: index to return
Returns:
data and label at that index
"""
files = self.files[index]
image = self._load_image(files["image"])
dem = self._load_image(files["dem"], shape=image.shape[1:])
image = torch.cat(tensors=[image, dem], dim=0)
sample = {"image": image}
if self.split == "train":
mask = self._load_target(files["target"])
sample["mask"] = mask
if self.transforms is not None:
sample = self.transforms(sample)
return sample
[docs] def __len__(self) -> int:
"""Return the number of data points in the dataset.
Returns:
length of the dataset
"""
return len(self.files)
def _load_files(self) -> list[dict[str, str]]:
"""Return the paths of the files in the dataset.
Returns:
list of dicts containing paths for each pair of image/dem/mask
"""
directory = os.path.join(self.root, self.metadata[self.split]["directory"])
images = glob.glob(
os.path.join(directory, "**", self.image_root, "*.tif"), recursive=True
)
files = []
for image in sorted(images):
dem = image.replace(self.image_root, self.dem_root)
dem = f"{os.path.splitext(dem)[0]}_RGEALTI.tif"
if self.split == "train":
target = image.replace(self.image_root, self.target_root)
target = f"{os.path.splitext(target)[0]}_UA2012.tif"
files.append(dict(image=image, dem=dem, target=target))
else:
files.append(dict(image=image, dem=dem))
return files
def _load_image(self, path: str, shape: Sequence[int] | None = None) -> Tensor:
"""Load a single image.
Args:
path: path to the image
shape: the (h, w) to resample the image to
Returns:
the image
"""
with rasterio.open(path) as f:
array: "np.typing.NDArray[np.float_]" = f.read(
out_shape=shape, out_dtype="float32", resampling=Resampling.bilinear
)
tensor = torch.from_numpy(array)
return tensor
def _load_target(self, path: str) -> Tensor:
"""Load the target mask for a single image.
Args:
path: path to the image
Returns:
the target mask
"""
with rasterio.open(path) as f:
array: "np.typing.NDArray[np.int_]" = f.read(
indexes=1, out_dtype="int32", resampling=Resampling.bilinear
)
tensor = torch.from_numpy(array)
tensor = tensor.to(torch.long)
return tensor
def _verify(self) -> None:
"""Verify the integrity of the dataset."""
# Check if the files already exist
exists = []
for split_info in self.metadata.values():
exists.append(
os.path.exists(os.path.join(self.root, split_info["directory"]))
)
if all(exists):
return
# Check if .zip files already exists (if so then extract)
exists = []
for split_info in self.metadata.values():
filepath = os.path.join(self.root, split_info["filename"])
if os.path.isfile(filepath):
if self.checksum and not check_integrity(filepath, split_info["md5"]):
raise RuntimeError("Dataset found, but corrupted.")
exists.append(True)
extract_archive(filepath)
else:
exists.append(False)
if all(exists):
return
raise DatasetNotFoundError(self)
[docs] def plot(
self,
sample: dict[str, Tensor],
show_titles: bool = True,
suptitle: str | None = None,
) -> Figure:
"""Plot a sample from the dataset.
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
"""
ncols = 2
image = sample["image"][:3]
image = image.to(torch.uint8)
image = image.permute(1, 2, 0).numpy()
dem = sample["image"][-1].numpy()
dem = percentile_normalization(dem, lower=0, upper=100, axis=(0, 1))
showing_mask = "mask" in sample
showing_prediction = "prediction" in sample
cmap = colors.ListedColormap(self.colormap)
if showing_mask:
mask = sample["mask"].numpy()
ncols += 1
if showing_prediction:
pred = sample["prediction"].numpy()
ncols += 1
fig, axs = plt.subplots(nrows=1, ncols=ncols, figsize=(10, ncols * 10))
axs[0].imshow(image)
axs[0].axis("off")
axs[1].imshow(dem)
axs[1].axis("off")
if showing_mask:
axs[2].imshow(mask, cmap=cmap, interpolation="none")
axs[2].axis("off")
if showing_prediction:
axs[3].imshow(pred, cmap=cmap, interpolation="none")
axs[3].axis("off")
elif showing_prediction:
axs[2].imshow(pred, cmap=cmap, interpolation="none")
axs[2].axis("off")
if show_titles:
axs[0].set_title("Image")
axs[1].set_title("DEM")
if showing_mask:
axs[2].set_title("Ground Truth")
if showing_prediction:
axs[3].set_title("Predictions")
elif showing_prediction:
axs[2].set_title("Predictions")
if suptitle is not None:
plt.suptitle(suptitle)
return fig
