Source code for torchgeo.datasets.seco
# Copyright (c) Microsoft Corporation. All rights reserved.
# Licensed under the MIT License.
"""Sentinel 2 imagery from the Seasonal Contrast paper."""
import os
import random
from typing import Callable, Dict, List, Optional
import matplotlib.pyplot as plt
import numpy as np
import rasterio
import torch
from PIL import Image
from torch import Tensor
from .geo import NonGeoDataset
from .utils import download_url, extract_archive, percentile_normalization
[docs]class SeasonalContrastS2(NonGeoDataset):
"""Sentinel 2 imagery from the Seasonal Contrast paper.
The `Seasonal Contrast imagery <https://github.com/ServiceNow/seasonal-contrast>`_
dataset contains Sentinel 2 imagery patches sampled from different points in time
around the 10k most populated cities on Earth.
Dataset features:
* Two versions: 100K and 1M patches
* 12 band Sentinel 2 imagery from 5 points in time at each location
If you use this dataset in your research, please cite the following paper:
* https://arxiv.org/pdf/2103.16607.pdf
"""
all_bands = [
"B1",
"B2",
"B3",
"B4",
"B5",
"B6",
"B7",
"B8",
"B8A",
"B9",
"B11",
"B12",
]
rgb_bands = ["B4", "B3", "B2"]
metadata = {
"100k": {
"url": "https://zenodo.org/record/4728033/files/seco_100k.zip?download=1",
"md5": "ebf2d5e03adc6e657f9a69a20ad863e0",
"filename": "seco_100k.zip",
"directory": "seasonal_contrast_100k",
},
"1m": {
"url": "https://zenodo.org/record/4728033/files/seco_1m.zip?download=1",
"md5": "187963d852d4d3ce6637743ec3a4bd9e",
"filename": "seco_1m.zip",
"directory": "seasonal_contrast_1m",
},
}
[docs] def __init__(
self,
root: str = "data",
version: str = "100k",
seasons: int = 1,
bands: List[str] = rgb_bands,
transforms: Optional[Callable[[Dict[str, Tensor]], Dict[str, Tensor]]] = None,
download: bool = False,
checksum: bool = False,
) -> None:
"""Initialize a new SeasonalContrastS2 instance.
.. versionadded:: 0.5
The *seasons* parameter.
Args:
root: root directory where dataset can be found
version: one of "100k" or "1m" for the version of the dataset to use
seasons: number of seasonal patches to sample per location, 1--5
bands: list of bands to load
transforms: a function/transform that takes input sample and its target as
entry and returns a transformed version
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 ``version`` argument is invalid
RuntimeError: if ``download=False`` and data is not found, or checksums
don't match
"""
assert version in self.metadata.keys()
assert seasons in range(5)
for band in bands:
assert band in self.all_bands
self.root = root
self.version = version
self.seasons = seasons
self.bands = bands
self.transforms = transforms
self.download = download
self.checksum = checksum
self._verify()
[docs] def __getitem__(self, index: int) -> Dict[str, Tensor]:
"""Return an index within the dataset.
Args:
index: index to return
Returns:
sample with an "image" in SCxHxW format where S is the number of seasons
.. versionchanged:: 0.5
Image shape changed from 5xCxHxW to SCxHxW
"""
root = os.path.join(
self.root, self.metadata[self.version]["directory"], f"{index:06}"
)
subdirs = os.listdir(root)
subdirs = random.sample(subdirs, self.seasons)
images = [self._load_patch(root, subdir) for subdir in subdirs]
sample = {"image": torch.cat(images)}
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 (10**5 if self.version == "100k" else 10**6) // 5
def _load_patch(self, root: str, subdir: str) -> Tensor:
"""Load a single image patch.
Args:
root: root directory containing all seasons
subdir: season to load
Returns:
the image with the subset of bands specified by ``self.bands``
"""
all_data = []
for band in self.bands:
fn = os.path.join(root, subdir, f"{band}.tif")
with rasterio.open(fn) as f:
band_data = f.read(1).astype(np.float32)
height, width = band_data.shape
size = min(height, width)
if size < 264:
# TODO: PIL resize is much slower than cv2, we should check to see
# what could be sped up throughout later. There is also a potential
# slowdown here from converting to/from a PIL Image just to resize.
# https://gist.github.com/calebrob6/748045ac8d844154067b2eefa47de92f
pil_image = Image.fromarray(band_data)
# Moved in PIL 9.1.0
try:
resample = Image.Resampling.BILINEAR
except AttributeError:
resample = Image.BILINEAR
band_data = np.array(
pil_image.resize((264, 264), resample=resample)
)
all_data.append(band_data)
image = torch.from_numpy(np.stack(all_data, axis=0))
return image
def _verify(self) -> None:
"""Verify the integrity of the dataset.
Raises:
RuntimeError: if ``download=False`` but dataset is missing or checksum fails
"""
# Check if the extracted files already exist
directory_path = os.path.join(
self.root, self.metadata[self.version]["directory"]
)
if os.path.exists(directory_path):
return
# Check if the zip files have already been downloaded
zip_path = os.path.join(self.root, self.metadata[self.version]["filename"])
if os.path.exists(zip_path):
self._extract()
return
# Check if the user requested to download the dataset
if not self.download:
raise RuntimeError(
f"Dataset not found in `root={self.root}` and `download=False`, "
"either specify a different `root` directory or use `download=True` "
"to automatically download the dataset."
)
# Download the dataset
self._download()
self._extract()
def _download(self) -> None:
"""Download the dataset."""
download_url(
self.metadata[self.version]["url"],
self.root,
filename=self.metadata[self.version]["filename"],
md5=self.metadata[self.version]["md5"] if self.checksum else None,
)
def _extract(self) -> None:
"""Extract the dataset."""
extract_archive(
os.path.join(self.root, self.metadata[self.version]["filename"])
)
[docs] def plot(
self,
sample: Dict[str, Tensor],
show_titles: bool = True,
suptitle: Optional[str] = None,
) -> plt.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 the RGB bands are included in ``self.bands`` or the sample
contains a "prediction" key
.. versionadded:: 0.2
"""
if "prediction" in sample:
raise ValueError("This dataset doesn't support plotting predictions")
rgb_indices = []
for band in self.rgb_bands:
if band in self.bands:
rgb_indices.append(self.bands.index(band))
else:
raise ValueError("Dataset doesn't contain some of the RGB bands")
fig, axes = plt.subplots(ncols=self.seasons, figsize=(20, 4))
if self.seasons == 1:
axes = [axes]
indices = torch.tensor(rgb_indices)
for i in range(self.seasons):
image = sample["image"][indices + i * len(self.bands)].numpy()
image = np.rollaxis(image, 0, 3)
image = percentile_normalization(image, 0, 100)
axes[i].imshow(image)
axes[i].axis("off")
if suptitle is not None:
plt.suptitle(suptitle)
return fig
