Source code for torchgeo.datasets.so2sat
# Copyright (c) Microsoft Corporation. All rights reserved.
# Licensed under the MIT License.
"""So2Sat dataset."""
import os
from collections.abc import Sequence
from typing import Callable, Optional, cast
import matplotlib.pyplot as plt
import numpy as np
import torch
from matplotlib.figure import Figure
from torch import Tensor
from .geo import NonGeoDataset
from .utils import check_integrity, percentile_normalization
[docs]class So2Sat(NonGeoDataset):
"""So2Sat dataset.
The `So2Sat <https://doi.org/10.1109/MGRS.2020.2964708>`__ dataset consists of
corresponding synthetic aperture radar and multispectral optical image data
acquired by the Sentinel-1 and Sentinel-2 remote sensing satellites, and a
corresponding local climate zones (LCZ) label. The dataset is distributed over
42 cities across different continents and cultural regions of the world, and comes
with a variety of different splits.
This implementation covers the *2nd* and *3rd* versions of the dataset as described
in the author's github repository: https://github.com/zhu-xlab/So2Sat-LCZ42.
The different versions are as follows:
Version 2: This version contains imagery from 52 cities and is split into train/val/test as follows:
* Training: 42 cities around the world
* Validation: western half of 10 other cities covering 10 cultural zones
* Testing: eastern half of the 10 other cities
Version 3: A version of the dataset with 3 different train/test splits, as follows:
* Random split: every city 80% training / 20% testing (randomly sampled)
* Block split: every city is split in a geospatial 80%/20%-manner
* Cultural 10: 10 cities from different cultural zones are held back for testing purposes
Dataset classes:
0. Compact high rise
1. Compact middle rise
2. Compact low rise
3. Open high rise
4. Open mid rise
5. Open low rise
6. Lightweight low rise
7. Large low rise
8. Sparsely built
9. Heavy industry
10. Dense trees
11. Scattered trees
12. Bush, scrub
13. Low plants
14. Bare rock or paved
15. Bare soil or sand
16. Water
If you use this dataset in your research, please cite the following paper:
* https://doi.org/10.1109/MGRS.2020.2964708
.. note::
The version 2 dataset can be automatically downloaded using the following bash
script:
.. code-block:: bash
for split in training validation testing
do
wget ftp://m1483140:m1483140@dataserv.ub.tum.de/$split.h5
done
or manually downloaded from https://dataserv.ub.tum.de/index.php/s/m1483140
This download will likely take several hours.
The version 3 datasets can be downloaded using the following bash script:
.. code-block:: bash
for version in random block culture_10
do
for split in training testing
do
wget -P $version/ ftp://m1613658:m1613658@dataserv.ub.tum.de/$version/$split.h5
done
done
or manually downloaded from https://mediatum.ub.tum.de/1613658
""" # noqa: E501
versions = ["2", "3_random", "3_block", "3_culture_10"]
filenames_by_version = {
"2": {
"train": "training.h5",
"validation": "validation.h5",
"test": "testing.h5",
},
"3_random": {"train": "random/training.h5", "test": "random/testing.h5"},
"3_block": {"train": "block/training.h5", "test": "block/testing.h5"},
"3_culture_10": {
"train": "culture_10/training.h5",
"test": "culture_10/testing.h5",
},
}
md5s_by_version = {
"2": {
"train": "702bc6a9368ebff4542d791e53469244",
"validation": "71cfa6795de3e22207229d06d6f8775d",
"test": "e81426102b488623a723beab52b31a8a",
},
"3_random": {
"train": "94e2e2e667b406c2adf61e113b42204e",
"test": "1e15c425585ce816342d1cd779d453d8",
},
"3_block": {
"train": "a91d6150e8b059dac86105853f377a11",
"test": "6414af1ec33ace417e879f9c88066d47",
},
"3_culture_10": {
"train": "702bc6a9368ebff4542d791e53469244",
"test": "58335ce34ca3a18424e19da84f2832fc",
},
}
classes = [
"Compact high rise",
"Compact mid rise",
"Compact low rise",
"Open high rise",
"Open mid rise",
"Open low rise",
"Lightweight low rise",
"Large low rise",
"Sparsely built",
"Heavy industry",
"Dense trees",
"Scattered trees",
"Bush, scrub",
"Low plants",
"Bare rock or paved",
"Bare soil or sand",
"Water",
]
all_s1_band_names = (
"S1_B1",
"S1_B2",
"S1_B3",
"S1_B4",
"S1_B5",
"S1_B6",
"S1_B7",
"S1_B8",
)
all_s2_band_names = (
"S2_B02",
"S2_B03",
"S2_B04",
"S2_B05",
"S2_B06",
"S2_B07",
"S2_B08",
"S2_B8A",
"S2_B11",
"S2_B12",
)
all_band_names = all_s1_band_names + all_s2_band_names
rgb_bands = ["S2_B04", "S2_B03", "S2_B02"]
BAND_SETS = {
"all": all_band_names,
"s1": all_s1_band_names,
"s2": all_s2_band_names,
"rgb": rgb_bands,
}
[docs] def __init__(
self,
root: str = "data",
version: str = "2",
split: str = "train",
bands: Sequence[str] = BAND_SETS["all"],
transforms: Optional[Callable[[dict[str, Tensor]], dict[str, Tensor]]] = None,
checksum: bool = False,
) -> None:
"""Initialize a new So2Sat dataset instance.
Args:
root: root directory where dataset can be found
version: one of "2", "3_random", "3_block", or "3_culture_10"
split: one of "train", "validation", or "test"
bands: a sequence of band names to use where the indices correspond to the
array index of combined Sentinel 1 and Sentinel 2
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`` argument is invalid
RuntimeError: if data is not found in ``root``, or checksums don't match
.. versionadded:: 0.3
The *bands* parameter.
.. versionadded:: 0.5
The *version* parameter.
"""
try:
import h5py # noqa: F401
except ImportError:
raise ImportError(
"h5py is not installed and is required to use this dataset"
)
assert version in self.versions
assert split in self.filenames_by_version[version]
self._validate_bands(bands)
self.s1_band_indices: "np.typing.NDArray[np.int_]" = np.array(
[
self.all_s1_band_names.index(b)
for b in bands
if b in self.all_s1_band_names
]
).astype(int)
self.s1_band_names = [self.all_s1_band_names[i] for i in self.s1_band_indices]
self.s2_band_indices: "np.typing.NDArray[np.int_]" = np.array(
[
self.all_s2_band_names.index(b)
for b in bands
if b in self.all_s2_band_names
]
).astype(int)
self.s2_band_names = [self.all_s2_band_names[i] for i in self.s2_band_indices]
self.bands = bands
self.root = root
self.version = version
self.split = split
self.transforms = transforms
self.checksum = checksum
self.fn = os.path.join(self.root, self.filenames_by_version[version][split])
if not self._check_integrity():
raise RuntimeError("Dataset not found or corrupted.")
with h5py.File(self.fn, "r") as f:
self.size: int = f["label"].shape[0]
[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
"""
import h5py
with h5py.File(self.fn, "r") as f:
s1 = f["sen1"][index].astype(np.float64) # convert from <f8 to float64
s1 = np.take(s1, indices=self.s1_band_indices, axis=2)
s2 = f["sen2"][index].astype(np.float64) # convert from <f8 to float64
s2 = np.take(s2, indices=self.s2_band_indices, axis=2)
# convert one-hot encoding to int64 then torch int
label = torch.tensor(f["label"][index].argmax())
s1 = np.rollaxis(s1, 2, 0) # convert to CxHxW format
s2 = np.rollaxis(s2, 2, 0) # convert to CxHxW format
s1 = torch.from_numpy(s1)
s2 = torch.from_numpy(s2)
sample = {"image": torch.cat([s1, s2]).float(), "label": label}
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 self.size
def _check_integrity(self) -> bool:
"""Check integrity of dataset.
Returns:
True if dataset files are found and/or MD5s match, else False
"""
md5 = self.md5s_by_version[self.version][self.split]
if not check_integrity(self.fn, md5 if self.checksum else None):
return False
return True
def _validate_bands(self, bands: Sequence[str]) -> None:
"""Validate list of bands.
Args:
bands: user-provided sequence of bands to load
Raises:
AssertionError: if ``bands`` is not a sequence
ValueError: if an invalid band name is provided
.. versionadded:: 0.3
"""
assert isinstance(bands, Sequence), "'bands' must be a sequence"
for band in bands:
if band not in self.all_band_names:
raise ValueError(f"'{band}' is an invalid band name.")
[docs] def plot(
self,
sample: dict[str, Tensor],
show_titles: bool = True,
suptitle: Optional[str] = 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
Raises:
ValueError: if RGB bands are not found in dataset
.. versionadded:: 0.2
"""
rgb_indices = []
for band in self.rgb_bands:
if band in self.s2_band_names:
idx = self.s2_band_names.index(band) + len(self.s1_band_names)
rgb_indices.append(idx)
else:
raise ValueError("Dataset doesn't contain some of the RGB bands")
image = np.take(sample["image"].numpy(), indices=rgb_indices, axis=0)
image = np.rollaxis(image, 0, 3)
image = percentile_normalization(image, 0, 100)
label = cast(int, sample["label"].item())
label_class = self.classes[label]
showing_predictions = "prediction" in sample
if showing_predictions:
prediction = cast(int, sample["prediction"].item())
prediction_class = self.classes[prediction]
fig, ax = plt.subplots(figsize=(4, 4))
ax.imshow(image)
ax.axis("off")
if show_titles:
title = f"Label: {label_class}"
if showing_predictions:
title += f"\nPrediction: {prediction_class}"
ax.set_title(title)
if suptitle is not None:
plt.suptitle(suptitle)
return fig