Source code for torchgeo.datasets.fair1m
# Copyright (c) Microsoft Corporation. All rights reserved.
# Licensed under the MIT License.
"""FAIR1M dataset."""
import glob
import os
from typing import Any, Callable, Dict, List, Optional, Tuple, cast
from xml.etree.ElementTree import Element, parse
import matplotlib.patches as patches
import matplotlib.pyplot as plt
import numpy as np
import torch
from PIL import Image
from torch import Tensor
from .geo import VisionDataset
from .utils import check_integrity, extract_archive
def parse_pascal_voc(path: str) -> Dict[str, Any]:
"""Read a PASCAL VOC annotation file.
Args:
path: path to xml file
Returns:
dict of image filename, points, and class labels
"""
et = parse(path)
element = et.getroot()
source = cast(Element, element.find("source"))
filename = cast(Element, source.find("filename")).text
labels, points = [], []
objects = cast(Element, element.find("objects"))
for obj in objects.findall("object"):
elm_points = cast(Element, obj.find("points"))
lis_points = elm_points.findall("point")
str_points = []
for point in lis_points:
text = cast(str, point.text)
str_points.append(text.split(","))
tup_points = [(float(p1), float(p2)) for p1, p2 in str_points]
possibleresult = cast(Element, obj.find("possibleresult"))
name = cast(Element, possibleresult.find("name"))
label = name.text
labels.append(label)
points.append(tup_points)
return dict(filename=filename, points=points, labels=labels)
class FAIR1M(VisionDataset):
"""FAIR1M dataset.
The `FAIR1M <http://gaofen-challenge.com/benchmark>`_
dataset is a dataset for remote sensing fine-grained oriented object detection.
Dataset features:
* 15,000+ images with 0.3-0.8 m per pixel resolution (1,000-10,000 px)
* 1 million object instances
* 5 object categories, 37 object sub-categories
* three spectral bands - RGB
* images taken by Gaofen satellites and Google Earth
Dataset format:
* images are three-channel tiffs
* labels are xml files with PASCAL VOC like annotations
Dataset classes:
0. Passenger Ship
1. Motorboat
2. Fishing Boat
3. Tugboat
4. other-ship
5. Engineering Ship
6. Liquid Cargo Ship
7. Dry Cargo Ship
8. Warship
9. Small Car
10. Bus
11. Cargo Truck
12. Dump Truck
13. other-vehicle
14. Van
15. Trailer
16. Tractor
17. Excavator
18. Truck Tractor
19. Boeing737
20. Boeing747
21. Boeing777
22. Boeing787
23. ARJ21
24. C919
25. A220
26. A321
27. A330
28. A350
29. other-airplane
30. Baseball Field
31. Basketball Court
32. Football Field
33. Tennis Court
34. Roundabout
35. Intersection
36. Bridge
If you use this dataset in your research, please cite the following paper:
* https://arxiv.org/abs/2103.05569
.. versionadded:: 0.2
"""
classes = {
"Passenger Ship": {"id": 0, "category": "Ship"},
"Motorboat": {"id": 1, "category": "Ship"},
"Fishing Boat": {"id": 2, "category": "Ship"},
"Tugboat": {"id": 3, "category": "Ship"},
"other-ship": {"id": 4, "category": "Ship"},
"Engineering Ship": {"id": 5, "category": "Ship"},
"Liquid Cargo Ship": {"id": 6, "category": "Ship"},
"Dry Cargo Ship": {"id": 7, "category": "Ship"},
"Warship": {"id": 8, "category": "Ship"},
"Small Car": {"id": 9, "category": "Vehicle"},
"Bus": {"id": 10, "category": "Vehicle"},
"Cargo Truck": {"id": 11, "category": "Vehicle"},
"Dump Truck": {"id": 12, "category": "Vehicle"},
"other-vehicle": {"id": 13, "category": "Vehicle"},
"Van": {"id": 14, "category": "Vehicle"},
"Trailer": {"id": 15, "category": "Vehicle"},
"Tractor": {"id": 16, "category": "Vehicle"},
"Excavator": {"id": 17, "category": "Vehicle"},
"Truck Tractor": {"id": 18, "category": "Vehicle"},
"Boeing737": {"id": 19, "category": "Airplane"},
"Boeing747": {"id": 20, "category": "Airplane"},
"Boeing777": {"id": 21, "category": "Airplane"},
"Boeing787": {"id": 22, "category": "Airplane"},
"ARJ21": {"id": 23, "category": "Airplane"},
"C919": {"id": 24, "category": "Airplane"},
"A220": {"id": 25, "category": "Airplane"},
"A321": {"id": 26, "category": "Airplane"},
"A330": {"id": 27, "category": "Airplane"},
"A350": {"id": 28, "category": "Airplane"},
"other-airplane": {"id": 29, "category": "Airplane"},
"Baseball Field": {"id": 30, "category": "Court"},
"Basketball Court": {"id": 31, "category": "Court"},
"Football Field": {"id": 32, "category": "Court"},
"Tennis Court": {"id": 33, "category": "Court"},
"Roundabout": {"id": 34, "category": "Road"},
"Intersection": {"id": 35, "category": "Road"},
"Bridge": {"id": 36, "category": "Road"},
}
image_root: str = "images"
labels_root: str = "labelXmls"
filenames = ["images.zip", "labelXmls.zip"]
md5s = ["a460fe6b1b5b276bf856ce9ac72d6568", "ca8666dc43a553f8d65e5dc671a8ac3c"]
[docs] def __init__(
self,
root: str = "data",
transforms: Optional[Callable[[Dict[str, Tensor]], Dict[str, Tensor]]] = None,
checksum: bool = False,
) -> None:
"""Initialize a new FAIR1M dataset instance.
Args:
root: root directory where dataset can be found
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)
"""
self.root = root
self.transforms = transforms
self.checksum = checksum
self._verify()
self.files = sorted(
glob.glob(os.path.join(self.root, self.labels_root, "*.xml"))
)
[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
"""
path = self.files[index]
parsed = parse_pascal_voc(path)
image = self._load_image(parsed["filename"])
boxes, labels = self._load_target(parsed["points"], parsed["labels"])
sample = {"image": image, "boxes": boxes, "label": labels}
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_image(self, path: str) -> Tensor:
"""Load a single image.
Args:
path: path to image
Returns:
the image
"""
path = os.path.join(self.root, self.image_root, path)
with Image.open(path) as img:
array: "np.typing.NDArray[np.int_]" = np.array(img.convert("RGB"))
tensor = torch.from_numpy(array)
# Convert from HxWxC to CxHxW
tensor = tensor.permute((2, 0, 1))
return tensor
def _load_target(
self, points: List[List[Tuple[float, float]]], labels: List[str]
) -> Tuple[Tensor, Tensor]:
"""Load the target mask for a single image.
Args:
points: list of point tuple lists
labels: list of class labels
Returns:
the target bounding boxes and labels
"""
labels_list = [self.classes[label]["id"] for label in labels]
boxes = torch.tensor(points).to(torch.float)
labels_tensor = torch.tensor(labels_list)
return boxes, labels_tensor
def _verify(self) -> None:
"""Verify the integrity of the dataset.
Raises:
RuntimeError: if checksum fails or the dataset is not found
"""
# Check if the files already exist
exists = []
for directory in [self.image_root, self.labels_root]:
exists.append(os.path.exists(os.path.join(self.root, directory)))
if all(exists):
return
# Check if .zip files already exists (if so extract)
exists = []
for filename, md5 in zip(self.filenames, self.md5s):
filepath = os.path.join(self.root, filename)
if os.path.isfile(filepath):
if self.checksum and not check_integrity(filepath, md5):
raise RuntimeError("Dataset found, but corrupted.")
exists.append(True)
extract_archive(filepath)
else:
exists.append(False)
if all(exists):
return
raise RuntimeError(
"Dataset not found in `root` directory, "
"specify a different `root` directory."
)
[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
"""
image = sample["image"].permute((1, 2, 0)).numpy()
ncols = 1
if "prediction_boxes" in sample:
ncols += 1
fig, axs = plt.subplots(ncols=ncols, figsize=(ncols * 10, 10))
if ncols < 2:
axs = [axs]
axs[0].imshow(image)
axs[0].axis("off")
polygons = [
patches.Polygon(points, color="r", fill=False)
for points in sample["boxes"].numpy()
]
for polygon in polygons:
axs[0].add_patch(polygon)
if show_titles:
axs[0].set_title("Ground Truth")
if ncols > 1:
axs[1].imshow(image)
axs[1].axis("off")
polygons = [
patches.Polygon(points, color="r", fill=False)
for points in sample["prediction_boxes"].numpy()
]
for polygon in polygons:
axs[0].add_patch(polygon)
if show_titles:
axs[1].set_title("Predictions")
if suptitle is not None:
plt.suptitle(suptitle)
return fig
