diff --git a/pysegcnn/core/dataset.py b/pysegcnn/core/dataset.py
index 98a24283adfbbfc3d6292f028f2bdf7a4b348471..d09b713630f5520ad3dd9b4c73cf796fe51af08c 100644
--- a/pysegcnn/core/dataset.py
+++ b/pysegcnn/core/dataset.py
@@ -35,45 +35,76 @@ from pysegcnn.core.utils import (img2np, is_divisible, tile_topleft_corner,
LOGGER = logging.getLogger(__name__)
-# generic image dataset class
class ImageDataset(Dataset):
-
- # allowed keyword arguments and default values
- default_kwargs = {
-
- # which bands to use, if use_bands=[], use all available bands
- 'use_bands': [],
-
- # each scene is divided into (tile_size x tile_size) blocks
- # each of these blocks is treated as a single sample
- 'tile_size': None,
-
- # a pattern to match the ground truth file naming convention
- 'gt_pattern': '*gt.tif',
-
- # whether to chronologically sort the samples
- 'sort': False,
-
- # the random seed used to split dataset into training, validation and
- # test data
- 'seed': 0,
-
- # the transformations to apply to the original image
- # artificially increases the training data size
- 'transforms': [],
-
- # whether to pad the image to be evenly divisible in square tiles
- # of size (tile_size x tile_size)
- 'pad': False
-
- }
-
- def __init__(self, root_dir, **kwargs):
+ """Base class for multispectral image data.
+
+ Inheriting from `torch.utils.data.Dataset` to be compliant to the PyTorch
+ standard. Furthermore, using instances of `torch.utils.data.Dataset`
+ enables the use of the handy `torch.utils.data.DataLoader` class during
+ model training.
+
+ Parameters
+ ----------
+ root_dir : `str`
+ The root directory, path to the dataset.
+ use_bands : `list` [`str`], optional
+ A list of the spectral bands to use. The default is [].
+ tile_size : `int` or `None`, optional
+ The size of the tiles. If not `None`, each scene is divided into square
+ tiles of shape (tile_size, tile_size). The default is None.
+ pad : `bool`, optional
+ Whether to center pad the input image. Set ``pad`` = True, if the
+ images are not evenly divisible by the ``tile_size``. The image data is
+ padded with a constant padding value of zero. For each image, the
+ corresponding ground truth image is padded with a "no data" label.
+ The default is False.
+ gt_pattern : `str`, optional
+ A pattern to match the ground truth naming convention. All directories
+ and subdirectories in ``root_dir`` are searched for files matching
+ ``gt_pattern``. The default is '*gt.tif'.
+ sort : `bool`, optional
+ Whether to chronologically sort the samples. Useful for time series
+ data. The default is False.
+ seed : `int`, optional
+ The random seed. Used to split the dataset into training, validation
+ and test set. Useful for reproducibility. The default is 0.
+ transforms : `list` [`pysegcnn.core.split.Augment`], optional
+ List of `pysegcnn.core.split.Augment` instances. Each item in
+ ``transforms`` generates a distinct transformed version of the dataset.
+ The total dataset is composed of the original untransformed dataset
+ together with each transformed version of it.
+ If ``transforms`` = [], only the original dataset is used.
+ The default is [].
+
+ Returns
+ -------
+ None.
+
+ """
+
+ def __init__(self, root_dir, use_bands=[], tile_size=None, pad=False,
+ gt_pattern='*gt.tif', sort=False, seed=0, transforms=[]):
super().__init__()
- # the root directory: path to the image dataset
+ # dataset configuration
self.root = root_dir
+ self.use_bands = use_bands
+ self.tile_size = tile_size
+ self.pad = pad
+ self.gt_pattern = gt_pattern
+ self.sort = sort
+ self.seed = seed
+ self.transforms = transforms
+
+ # initialize instance attributes
+ self._init_attributes()
+ # the samples of the dataset
+ self.scenes = self.compose_scenes()
+ self._assert_compose_scenes()
+
+ def _init_attributes(self):
+ """Initialize the class instance attributes."""
# the size of a scene/patch in the dataset
self.size = self.get_size()
self._assert_get_size()
@@ -88,26 +119,6 @@ class ImageDataset(Dataset):
self._assert_get_labels()
self.labels = self._build_labels()
- # initialize keyword arguments
- self._init_kwargs(**kwargs)
-
- # the samples of the dataset
- self.scenes = self.compose_scenes()
- self._assert_compose_scenes()
-
- def _init_kwargs(self, **kwargs):
-
- # check if the keyword arguments are correctly specified
- if not set(kwargs.keys()).issubset(self.default_kwargs.keys()):
- raise TypeError('Valid keyword arguments are: \n' +
- '\n'.join('- {}'.format(k) for k in
- self.default_kwargs.keys()))
-
- # update default arguments with specified keyword argument values
- self.default_kwargs.update(kwargs)
- for k, v in self.default_kwargs.items():
- setattr(self, k, v)
-
# check which bands to use
self.use_bands = (self.use_bands if self.use_bands else
[*self.bands.values()])
@@ -140,12 +151,25 @@ class ImageDataset(Dataset):
.format(self.cval))
def _build_labels(self):
+ """Build the label dictionary.
+
+ Returns
+ -------
+ labels : `dict` [`int`, `dict`]
+ The label dictionary. The keys are the values of the class labels
+ in the ground truth ``y``. Each nested `dict` should have keys:
+ ``'color'``
+ A named color (`str`).
+ ``'label'``
+ The name of the class label (`str`).
+
+ """
return {band.id: {'label': band.name.replace('_', ' '),
'color': band.color}
for band in self._label_class}
def _assert_compose_scenes(self):
-
+ """Check whether compose_scenes() is correctly implemented."""
# list of required keys
self.keys = self.use_bands + ['gt', 'date', 'tile', 'transform', 'id']
@@ -163,12 +187,14 @@ class ImageDataset(Dataset):
.format(self.keys))
def _assert_get_size(self):
+ """Check whether get_size() is correctly implemented."""
if not isinstance(self.size, tuple) and len(self.size) == 2:
raise TypeError('{}.get_size() should return the spatial size of '
'an image sample as tuple, i.e. (height, width).'
.format(self.__class__.__name__))
def _assert_get_sensor(self):
+ """Check whether get_sensor() is correctly implemented."""
if not isinstance(self.sensor, enum.EnumMeta):
raise TypeError('{}.get_sensor() should return an instance of '
'enum.Enum, containing an enumeration of the '
@@ -178,6 +204,7 @@ class ImageDataset(Dataset):
.format(self.__class__.__name__))
def _assert_get_labels(self):
+ """Check whether get_labels() is correctly implemented."""
if not issubclass(self._label_class, Label):
raise TypeError('{}.get_labels() should return an instance of '
'pysegcnn.core.constants.Label, '
@@ -188,16 +215,32 @@ class ImageDataset(Dataset):
'pysegcnn.core.constants.py.'
.format(self.__class__.__name__))
- # the __len__() method returns the number of samples in the dataset
def __len__(self):
- # number of (tiles x channels x height x width) patches after each
- # scene is decomposed to tiles blocks
+ """Return the number of samples in the dataset.
+
+ Returns
+ -------
+ nsamples : `int`
+ The number of samples in the dataset.
+ """
return len(self.scenes)
- # the __getitem__() method returns a single sample of the dataset given an
- # index, i.e. an array/tensor of shape (channels x height x width)
def __getitem__(self, idx):
+ """Return the data of a sample of the dataset given an index ``idx``.
+
+ Parameters
+ ----------
+ idx : `int`
+ The index of the sample.
+
+ Returns
+ -------
+ x : `torch.Tensor`
+ The sample input data.
+ y : `torch.Tensor`
+ The sample ground truth.
+ """
# select a scene
scene = self.read_scene(idx)
@@ -209,88 +252,191 @@ class ImageDataset(Dataset):
# preprocess samples
x, y = self.preprocess(data, gt)
- # optional transformation
+ # apply transformation
if scene['transform'] is not None:
x, y = scene['transform'](x, y)
# convert to torch tensors
- x, y = self.to_tensor(x, y)
+ x = self.to_tensor(x, dtype=torch.float32)
+ y = self.to_tensor(y, dtype=torch.uint8)
return x, y
- # the compose_scenes() method has to be implemented by the class inheriting
- # the ImageDataset class
- # compose_scenes() should return a list of dictionaries, where each
- # dictionary represent one sample of the dataset, a scene or a tile
- # of a scene, etc.
- # the dictionaries should have the following (key, value) pairs:
- # - (band_1, path_to_band_1.tif)
- # - (band_2, path_to_band_2.tif)
- # - ...
- # - (band_n, path_to_band_n.tif)
- # - (gt, path_to_ground_truth.tif)
- # - (tile, None or int)
- # - ...
- def compose_scenes(self, *args, **kwargs):
+ def compose_scenes(self):
+ """Build the list of samples of the dataset.
+
+ Each sample is represented by a dictionary.
+
+ Raises
+ ------
+ NotImplementedError
+ Raised if the `pysegcnn.core.dataset.ImageDataset` class is not
+ inherited.
+
+ Returns
+ -------
+ samples : `list` [`dict`]
+ Each dictionary representing a sample should have keys:
+ ``'band_name_1'``
+ Path to the file of band_1.
+ ``'band_name_2'``
+ Path to the file of band_2.
+ ``'band_name_n'``
+ Path to the file of band_n.
+ ``'gt'``
+ Path to the ground truth file.
+ ``'date'``
+ The date of the sample.
+ ``'tile'``
+ The tile id of the sample.
+ ``'transform'``
+ The transformation to apply.
+ ``'id'``
+ The scene identifier.
+
+ """
raise NotImplementedError('Inherit the ImageDataset class and '
'implement the method.')
- # the get_size() method has to be implemented by the class inheriting
- # the ImageDataset class
- # get_size() method should return the image size as tuple, (height, width)
- def get_size(self, *args, **kwargs):
+ def get_size(self):
+ """Return the size of the images in the dataset.
+
+ Raises
+ ------
+ NotImplementedError
+ Raised if the `pysegcnn.core.dataset.ImageDataset` class is not
+ inherited.
+
+ Returns
+ -------
+ size : `tuple`
+ The image size (height, width).
+
+ """
raise NotImplementedError('Inherit the ImageDataset class and '
'implement the method.')
- # the get_sensor() method has to be implemented by the class inheriting
- # the ImageDataset class
- # get_sensor() should return an enum.Enum with the following
- # (name: str, value: int) tuples:
- # - (red, 2)
- # - (green, 3)
- # - ...
- def get_sensor(self, *args, **kwargs):
+ def get_sensor(self):
+ """Return an enumeration of the bands of the sensor of the dataset.
+
+ Examples can be found in `pysegcnn.core.constants`.
+
+ Raises
+ ------
+ NotImplementedError
+ Raised if the `pysegcnn.core.dataset.ImageDataset` class is not
+ inherited.
+
+ Returns
+ -------
+ sensor : `enum.Enum`
+ An enumeration of the bands of the sensor.
+
+ """
raise NotImplementedError('Inherit the ImageDataset class and '
'implement the method.')
- # the get_labels() method has to be implemented by the class inheriting
- # the ImageDataset class
- # get_labels() should return a dictionary with the following
- # (key: int, value: str) pairs:
- # - (0, label_1_name)
- # - (1, label_2_name)
- # - ...
- # - (n, label_n_name)
- # where the keys should be the values representing the values of the
- # corresponding label in the ground truth mask
- # the labels in the dictionary determine the classes to be segmented
- def get_labels(self, *args, **kwargs):
+ def get_labels(self):
+ """Return an enumeration of the class labels of the dataset.
+
+ Examples can be found in `pysegcnn.core.constants`.
+
+ Raises
+ ------
+ NotImplementedError
+ Raised if the `pysegcnn.core.dataset.ImageDataset` class is not
+ inherited.
+
+ Returns
+ -------
+ labels : `enum.Enum`
+ The class labels.
+
+ """
raise NotImplementedError('Inherit the ImageDataset class and '
'implement the method.')
- # the preprocess() method has to be implemented by the class inheriting
- # the ImageDataset class
- # preprocess() should return two torch.tensors:
- # - input data: tensor of shape (bands, height, width)
- # - ground truth: tensor of shape (height, width)
def preprocess(self, data, gt):
+ """Preprocess a sample before feeding it to a model.
+
+ Parameters
+ ----------
+ data : `numpy.ndarray`
+ The sample input data.
+ gt : `numpy.ndarray`
+ The sample ground truth.
+
+ Raises
+ ------
+ NotImplementedError
+ Raised if the `pysegcnn.core.dataset.ImageDataset` class is not
+ inherited.
+
+ Returns
+ -------
+ data : `numpy.ndarray`
+ The preprocessed input data.
+ gt : `numpy.ndarray`
+ The preprocessed ground truth data.
+
+ """
raise NotImplementedError('Inherit the ImageDataset class and '
'implement the method.')
- # the parse_scene_id() method has to be implemented by the class inheriting
- # the ImageDataset class
- # the input to the parse_scene_id() method is a string describing a scene
- # id, e.g. an id of a Landsat or a Sentinel scene
- # parse_scene_id() should return a dictionary containing the scene metadata
- def parse_scene_id(self, scene):
+ def parse_scene_id(self, scene_id):
+ """Parse the scene identifier.
+
+ Parameters
+ ----------
+ scene_id : `str`
+ A scene identifier.
+
+ Raises
+ ------
+ NotImplementedError
+ Raised if the `pysegcnn.core.dataset.ImageDataset` class is not
+ inherited.
+
+ Returns
+ -------
+ scene : `dict` or `None`
+ A dictionary containing scene metadata. If `None`, ``scene_id`` is
+ not a valid scene identifier.
+
+ """
raise NotImplementedError('Inherit the ImageDataset class and '
'implement the method.')
- # _read_scene() reads all the bands and the ground truth mask in a
- # scene/tile to a numpy array and returns a dictionary with
- # (key, value) = ('band_name', np.ndarray(band_data))
def read_scene(self, idx):
-
+ """Read the data of the sample with index ``idx``.
+
+ Parameters
+ ----------
+ idx : `int`
+ The index of the sample.
+
+ Returns
+ -------
+ scene_data : `dict`
+ The sample data dictionary with keys:
+ ``'band_name_1'``
+ data of band_1 (`numpy.ndarray`).
+ ``'band_name_2'``
+ data of band_2 (`numpy.ndarray`).
+ ``'band_name_n'``
+ data of band_n (`numpy.ndarray`).
+ ``'gt'``
+ data of the ground truth (`numpy.ndarray`).
+ ``'date'``
+ The date of the sample.
+ ``'tile'``
+ The tile id of the sample.
+ ``'transform'``
+ The transformation to apply.
+ ``'id'``
+ The scene identifier.
+
+ """
# select a scene from the root directory
scene = self.scenes[idx]
@@ -309,22 +455,71 @@ class ImageDataset(Dataset):
return scene_data
- # _build_samples() stacks all bands of a scene/tile into a
- # numpy array of shape (bands x height x width)
def build_samples(self, scene):
-
+ """Stack the bands of a sample in a single array.
+
+ Parameters
+ ----------
+ scene : `dict`
+ The sample data dictionary with keys:
+ ``'band_name_1'``
+ data of band_1 (`numpy.ndarray`).
+ ``'band_name_2'``
+ data of band_2 (`numpy.ndarray`).
+ ``'band_name_n'``
+ data of band_n (`numpy.ndarray`).
+ ``'gt'``
+ data of the ground truth (`numpy.ndarray`).
+ ``'date'``
+ The date of the sample.
+ ``'tile'``
+ The tile id of the sample.
+ ``'transform'``
+ The transformation to apply.
+ ``'id'``
+ The scene identifier.
+
+ Returns
+ -------
+ stack : `numpy.ndarray`
+ The input data of the sample.
+ gt : TYPE
+ The ground truth of the sample.
+
+ """
# iterate over the channels to stack
stack = np.stack([scene[band] for band in self.use_bands], axis=0)
gt = scene['gt']
return stack, gt
- def to_tensor(self, x, y):
- return (torch.tensor(x.copy(), dtype=torch.float32),
- torch.tensor(y.copy(), dtype=torch.uint8))
+ def to_tensor(self, x, dtype):
+ """Convert ``x`` to `torch.Tensor`.
+
+ Parameters
+ ----------
+ x : array_like
+ The input data.
+ dtype : `torch.dtype`
+ The data type used to convert ``x``.
+
+ Returns
+ -------
+ x : `torch.Tensor`
+ The input data tensor.
+
+ """
+ return torch.tensor(np.asarray(x).copy(), dtype=dtype)
def __repr__(self):
+ """Dataset representation.
+
+ Returns
+ -------
+ fs : `str`
+ Representation string.
+ """
# representation string to print
fs = self.__class__.__name__ + '(\n'
@@ -358,17 +553,106 @@ class ImageDataset(Dataset):
return fs
-
class StandardEoDataset(ImageDataset):
-
- def __init__(self, root_dir, **kwargs):
+ """Base class for standard Earth Observation style datasets.
+
+ `pysegcnn.core.dataset.StandardEoDataset` implements the
+ `~pysegcnn.core.dataset.StandardEoDataset.compose_scenes` method for
+ datasets with the following directory structure:
+
+ root_dir/
+ scene_id_1/
+ scene_id_1_B1.tif
+ scene_id_1_B2.tif
+ .
+ .
+ .
+ scene_id_1_BN.tif
+ scene_id_2/
+ scene_id_2_B1.tif
+ scene_id_2_B2.tif
+ .
+ .
+ .
+ scene_id_2_BN.tif
+ .
+ .
+ .
+ scene_id_N/
+ .
+ .
+ .
+
+ If your dataset shares this directory structure, you can directly inherit
+ `pysegcnn.core.dataset.StandardEoDataset` and implement the remaining
+ methods.
+
+ See `pysegcnn.core.dataset.SparcsDataset` for an example.
+
+ Parameters
+ ----------
+ root_dir : `str`
+ The root directory, path to the dataset.
+ use_bands : `list` [`str`], optional
+ A list of the spectral bands to use. The default is [].
+ tile_size : `int` or `None`, optional
+ The size of the tiles. If not `None`, each scene is divided into square
+ tiles of shape (tile_size, tile_size). The default is None.
+ pad : `bool`, optional
+ Whether to center pad the input image. Set ``pad`` = True, if the
+ images are not evenly divisible by the ``tile_size``. The image data is
+ padded with a constant padding value of zero. For each image, the
+ corresponding ground truth image is padded with a "no data" label.
+ The default is False.
+ gt_pattern : `str`, optional
+ A pattern to match the ground truth naming convention. All directories
+ and subdirectories in ``root_dir`` are searched for files matching
+ ``gt_pattern``. The default is '*gt.tif'.
+ sort : `bool`, optional
+ Whether to chronologically sort the samples. Useful for time series
+ data. The default is False.
+ seed : `int`, optional
+ The random seed. Used to split the dataset into training, validation
+ and test set. Useful for reproducibility. The default is 0.
+ transforms : `list` [`pysegcnn.core.split.Augment`], optional
+ List of `pysegcnn.core.split.Augment` instances. Each item in
+ ``transforms`` generates a distinct transformed version of the dataset.
+ The total dataset is composed of the original untransformed dataset
+ together with each transformed version of it.
+ If ``transforms`` = [], only the original dataset is used.
+ The default is [].
+
+ Returns
+ -------
+ None.
+
+ """
+
+ def __init__(self, root_dir, use_bands=[], tile_size=None, pad=False,
+ gt_pattern='*gt.tif', sort=False, seed=0, transforms=[]):
# initialize super class ImageDataset
- super().__init__(root_dir, **kwargs)
+ super().__init__(root_dir, use_bands, tile_size, pad, gt_pattern,
+ sort, seed, transforms)
+
+ def _get_band_number(self, path):
+ """Return the band number of a scene .tif file.
+
+ Parameters
+ ----------
+ path : `str`
+ The path to the .tif file.
+
+ Raises
+ ------
+ ValueError
+ Raised if ``path`` is not a .tif file.
- # returns the band number of a Landsat8 or Sentinel2 tif file
- # path: path to a tif file
- def get_band_number(self, path):
+ Returns
+ -------
+ band : `int` or `str`
+ The band number.
+ """
# check whether the path leads to a tif file
if not path.endswith(('tif', 'TIF')):
raise ValueError('Expected a path to a tif file.')
@@ -377,7 +661,7 @@ class StandardEoDataset(ImageDataset):
fname = os.path.basename(path)
# search for numbers following a "B" in the filename
- band = re.search('B\dA|B\d{1,2}', fname)[0].replace('B', '')
+ band = re.search('B\\dA|B\\d{1,2}', fname)[0].replace('B', '')
# try converting to an integer:
# raises a ValueError for Sentinel2 8A band
@@ -388,14 +672,34 @@ class StandardEoDataset(ImageDataset):
return band
- # store_bands() writes the paths to the data of each scene to a dictionary
- # only the bands of interest are stored
- def store_bands(self, bands, gt):
-
+ def _store_bands(self, bands, gt):
+ """Write the bands of interest to a dictionary.
+
+ Parameters
+ ----------
+ bands : `list` [`str`]
+ Paths to the .tif files of the bands of the scene.
+ gt : `str`
+ Path to the ground truth of the scene.
+
+ Returns
+ -------
+ scene_data : `dict`
+ The scene data dictionary with keys:
+ ``'band_name_1'``
+ Path to the .tif file of band_1.
+ ``'band_name_2'``
+ Path to the .tif file of band_2.
+ ``'band_name_n'``
+ Path to the .tif file of band_n.
+ ``'gt'``
+ Path to the ground truth file.
+
+ """
# store the bands of interest in a dictionary
scene_data = {}
for i, b in enumerate(bands):
- band = self.bands[self.get_band_number(b)]
+ band = self.bands[self._get_band_number(b)]
if band in self.use_bands:
scene_data[band] = b
@@ -404,12 +708,33 @@ class StandardEoDataset(ImageDataset):
return scene_data
- # compose_scenes() creates a list of dictionaries containing the paths
- # to the tif files of each scene
- # if the scenes are divided into tiles, each tile has its own entry
- # with corresponding tile id
def compose_scenes(self):
-
+ """Build the list of samples of the dataset.
+
+ Each sample is represented by a dictionary.
+
+ Returns
+ -------
+ scenes : `list` [`dict`]
+ Each item in ``scenes`` is a `dict` with keys:
+ ``'band_name_1'``
+ Path to the file of band_1.
+ ``'band_name_2'``
+ Path to the file of band_2.
+ ``'band_name_n'``
+ Path to the file of band_n.
+ ``'gt'``
+ Path to the ground truth file.
+ ``'date'``
+ The date of the sample.
+ ``'tile'``
+ The tile id of the sample.
+ ``'transform'``
+ The transformation to apply.
+ ``'id'``
+ The scene identifier.
+
+ """
# search the root directory
scenes = []
self.gt = []
@@ -447,7 +772,7 @@ class StandardEoDataset(ImageDataset):
for transf in self.transforms:
# store the bands and the ground truth mask of the tile
- data = self.store_bands(bands, gt)
+ data = self._store_bands(bands, gt)
# the name of the scene
data['id'] = scene['id']
@@ -472,122 +797,518 @@ class StandardEoDataset(ImageDataset):
return scenes
-# SparcsDataset class: inherits from the generic ImageDataset class
class SparcsDataset(StandardEoDataset):
-
- def __init__(self, root_dir, **kwargs):
+ """Dataset class for the `Sparcs`_ dataset.
+
+ .. _Sparcs:
+ https://www.usgs.gov/land-resources/nli/landsat/spatial-procedures-automated-removal-cloud-and-shadow-sparcs-validation
+
+ Parameters
+ ----------
+ root_dir : `str`
+ The root directory, path to the dataset.
+ use_bands : `list` [`str`], optional
+ A list of the spectral bands to use. The default is [].
+ tile_size : `int` or `None`, optional
+ The size of the tiles. If not `None`, each scene is divided into square
+ tiles of shape (tile_size, tile_size). The default is None.
+ pad : `bool`, optional
+ Whether to center pad the input image. Set ``pad`` = True, if the
+ images are not evenly divisible by the ``tile_size``. The image data is
+ padded with a constant padding value of zero. For each image, the
+ corresponding ground truth image is padded with a "no data" label.
+ The default is False.
+ gt_pattern : `str`, optional
+ A pattern to match the ground truth naming convention. All directories
+ and subdirectories in ``root_dir`` are searched for files matching
+ ``gt_pattern``. The default is '*gt.tif'.
+ sort : `bool`, optional
+ Whether to chronologically sort the samples. Useful for time series
+ data. The default is False.
+ seed : `int`, optional
+ The random seed. Used to split the dataset into training, validation
+ and test set. Useful for reproducibility. The default is 0.
+ transforms : `list` [`pysegcnn.core.split.Augment`], optional
+ List of `pysegcnn.core.split.Augment` instances. Each item in
+ ``transforms`` generates a distinct transformed version of the dataset.
+ The total dataset is composed of the original untransformed dataset
+ together with each transformed version of it.
+ If ``transforms`` = [], only the original dataset is used.
+ The default is [].
+
+ Returns
+ -------
+ None.
+
+ """
+
+ def __init__(self, root_dir, use_bands=[], tile_size=None, pad=False,
+ gt_pattern='*gt.tif', sort=False, seed=0, transforms=[]):
# initialize super class StandardEoDataset
- super().__init__(root_dir, **kwargs)
+ super().__init__(root_dir, use_bands, tile_size, pad, gt_pattern,
+ sort, seed, transforms)
- # image size of the Sparcs dataset: (height, width)
def get_size(self):
+ """Image size of the Sparcs dataset.
+
+ Returns
+ -------
+ size : `tuple`
+ The image size (height, width).
+
+ """
return (1000, 1000)
- # Landsat 8 bands of the Sparcs dataset
def get_sensor(self):
+ """Landsat 8 bands of the Sparcs dataset.
+
+ Returns
+ -------
+ sensor : `enum.Enum`
+ An enumeration of the bands of the sensor.
+
+ """
return Landsat8
- # class labels of the Sparcs dataset
def get_labels(self):
+ """Class labels of the Sparcs dataset.
+
+ Returns
+ -------
+ labels : `enum.Enum`
+ The class labels.
+
+ """
return SparcsLabels
- # preprocessing of the Sparcs dataset
def preprocess(self, data, gt):
-
+ """Preprocess Sparcs dataset images.
+
+ Parameters
+ ----------
+ data : `numpy.ndarray`
+ The sample input data.
+ gt : `numpy.ndarray`
+ The sample ground truth.
+
+ Returns
+ -------
+ data : `numpy.ndarray`
+ The preprocessed input data.
+ gt : `numpy.ndarray`
+ The preprocessed ground truth data.
+
+ """
# if the preprocessing is not done externally, implement it here
return data, gt
- # function that parses the date from a Landsat 8 scene id
- def parse_scene_id(self, scene):
- return parse_landsat_scene(scene)
+ def parse_scene_id(self, scene_id):
+ """Parse Sparcs scene identifiers (Landsat 8).
+ Parameters
+ ----------
+ scene_id : `str`
+ A scene identifier.
-class ProSnowDataset(StandardEoDataset):
+ Returns
+ -------
+ scene : `dict` or `None`
+ A dictionary containing scene metadata. If `None`, ``scene_id`` is
+ not a valid Landsat scene identifier.
+
+ """
+ return parse_landsat_scene(scene_id)
- def __init__(self, root_dir, **kwargs):
+
+class ProSnowDataset(StandardEoDataset):
+ """Dataset class for the ProSnow datasets.
+
+ Parameters
+ ----------
+ root_dir : `str`
+ The root directory, path to the dataset.
+ use_bands : `list` [`str`], optional
+ A list of the spectral bands to use. The default is [].
+ tile_size : `int` or `None`, optional
+ The size of the tiles. If not `None`, each scene is divided into square
+ tiles of shape (tile_size, tile_size). The default is None.
+ pad : `bool`, optional
+ Whether to center pad the input image. Set ``pad`` = True, if the
+ images are not evenly divisible by the ``tile_size``. The image data is
+ padded with a constant padding value of zero. For each image, the
+ corresponding ground truth image is padded with a "no data" label.
+ The default is False.
+ gt_pattern : `str`, optional
+ A pattern to match the ground truth naming convention. All directories
+ and subdirectories in ``root_dir`` are searched for files matching
+ ``gt_pattern``. The default is '*gt.tif'.
+ sort : `bool`, optional
+ Whether to chronologically sort the samples. Useful for time series
+ data. The default is False.
+ seed : `int`, optional
+ The random seed. Used to split the dataset into training, validation
+ and test set. Useful for reproducibility. The default is 0.
+ transforms : `list` [`pysegcnn.core.split.Augment`], optional
+ List of `pysegcnn.core.split.Augment` instances. Each item in
+ ``transforms`` generates a distinct transformed version of the dataset.
+ The total dataset is composed of the original untransformed dataset
+ together with each transformed version of it.
+ If ``transforms`` = [], only the original dataset is used.
+ The default is [].
+
+ Returns
+ -------
+ None.
+
+ """
+
+ def __init__(self, root_dir, use_bands=[], tile_size=None, pad=False,
+ gt_pattern='*gt.tif', sort=False, seed=0, transforms=[]):
# initialize super class StandardEoDataset
- super().__init__(root_dir, **kwargs)
+ super().__init__(root_dir, use_bands, tile_size, pad, gt_pattern,
+ sort, seed, transforms)
- # Sentinel 2 bands
def get_sensor(self):
+ """Sentinel 2 bands of the ProSnow datasets.
+
+ Returns
+ -------
+ sensor : `enum.Enum`
+ An enumeration of the bands of the sensor.
+
+ """
return Sentinel2
- # class labels of the ProSnow dataset
def get_labels(self):
+ """Class labels of the ProSnow datasets.
+
+ Returns
+ -------
+ labels : `enum.Enum`
+ The class labels.
+
+ """
return ProSnowLabels
- # preprocessing of the ProSnow dataset
def preprocess(self, data, gt):
-
+ """Preprocess ProSnow dataset images.
+
+ Parameters
+ ----------
+ data : `numpy.ndarray`
+ The sample input data.
+ gt : `numpy.ndarray`
+ The sample ground truth.
+
+ Returns
+ -------
+ data : `numpy.ndarray`
+ The preprocessed input data.
+ gt : `numpy.ndarray`
+ The preprocessed ground truth data.
+
+ """
# if the preprocessing is not done externally, implement it here
return data, gt
- # function that parses the date from a Sentinel 2 scene id
- def parse_scene_id(self, scene):
- return parse_sentinel2_scene(scene)
+ def parse_scene_id(self, scene_id):
+ """Parse ProSnow scene identifiers (Sentinel 2).
+ Parameters
+ ----------
+ scene_id : `str`
+ A scene identifier.
-class ProSnowGarmisch(ProSnowDataset):
+ Returns
+ -------
+ scene : `dict` or `None`
+ A dictionary containing scene metadata. If `None`, ``scene_id`` is
+ not a valid Sentinel-2 scene identifier.
- def __init__(self, root_dir, **kwargs):
- # initialize super class StandardEoDatasets
- super().__init__(root_dir, **kwargs)
+ """
+ return parse_sentinel2_scene(scene_id)
+
+
+class ProSnowGarmisch(ProSnowDataset):
+ """Dataset class for the ProSnow Garmisch dataset.
+
+ Parameters
+ ----------
+ root_dir : `str`
+ The root directory, path to the dataset.
+ use_bands : `list` [`str`], optional
+ A list of the spectral bands to use. The default is [].
+ tile_size : `int` or `None`, optional
+ The size of the tiles. If not `None`, each scene is divided into square
+ tiles of shape (tile_size, tile_size). The default is None.
+ pad : `bool`, optional
+ Whether to center pad the input image. Set ``pad`` = True, if the
+ images are not evenly divisible by the ``tile_size``. The image data is
+ padded with a constant padding value of zero. For each image, the
+ corresponding ground truth image is padded with a "no data" label.
+ The default is False.
+ gt_pattern : `str`, optional
+ A pattern to match the ground truth naming convention. All directories
+ and subdirectories in ``root_dir`` are searched for files matching
+ ``gt_pattern``. The default is '*gt.tif'.
+ sort : `bool`, optional
+ Whether to chronologically sort the samples. Useful for time series
+ data. The default is False.
+ seed : `int`, optional
+ The random seed. Used to split the dataset into training, validation
+ and test set. Useful for reproducibility. The default is 0.
+ transforms : `list` [`pysegcnn.core.split.Augment`], optional
+ List of `pysegcnn.core.split.Augment` instances. Each item in
+ ``transforms`` generates a distinct transformed version of the dataset.
+ The total dataset is composed of the original untransformed dataset
+ together with each transformed version of it.
+ If ``transforms`` = [], only the original dataset is used.
+ The default is [].
+
+ Returns
+ -------
+ None.
+
+ """
+
+ def __init__(self, root_dir, use_bands=[], tile_size=None, pad=False,
+ gt_pattern='*gt.tif', sort=False, seed=0, transforms=[]):
+ # initialize super class StandardEoDataset
+ super().__init__(root_dir, use_bands, tile_size, pad, gt_pattern,
+ sort, seed, transforms)
def get_size(self):
+ """Image size of the ProSnow Garmisch dataset.
+
+ Returns
+ -------
+ size : `tuple`
+ The image size (height, width).
+
+ """
return (615, 543)
class ProSnowObergurgl(ProSnowDataset):
-
- def __init__(self, root_dir, **kwargs):
+ """Dataset class for the ProSnow Obergurgl dataset.
+
+ Parameters
+ ----------
+ root_dir : `str`
+ The root directory, path to the dataset.
+ use_bands : `list` [`str`], optional
+ A list of the spectral bands to use. The default is [].
+ tile_size : `int` or `None`, optional
+ The size of the tiles. If not `None`, each scene is divided into square
+ tiles of shape (tile_size, tile_size). The default is None.
+ pad : `bool`, optional
+ Whether to center pad the input image. Set ``pad`` = True, if the
+ images are not evenly divisible by the ``tile_size``. The image data is
+ padded with a constant padding value of zero. For each image, the
+ corresponding ground truth image is padded with a "no data" label.
+ The default is False.
+ gt_pattern : `str`, optional
+ A pattern to match the ground truth naming convention. All directories
+ and subdirectories in ``root_dir`` are searched for files matching
+ ``gt_pattern``. The default is '*gt.tif'.
+ sort : `bool`, optional
+ Whether to chronologically sort the samples. Useful for time series
+ data. The default is False.
+ seed : `int`, optional
+ The random seed. Used to split the dataset into training, validation
+ and test set. Useful for reproducibility. The default is 0.
+ transforms : `list` [`pysegcnn.core.split.Augment`], optional
+ List of `pysegcnn.core.split.Augment` instances. Each item in
+ ``transforms`` generates a distinct transformed version of the dataset.
+ The total dataset is composed of the original untransformed dataset
+ together with each transformed version of it.
+ If ``transforms`` = [], only the original dataset is used.
+ The default is [].
+
+ Returns
+ -------
+ None.
+
+ """
+
+ def __init__(self, root_dir, use_bands=[], tile_size=None, pad=False,
+ gt_pattern='*gt.tif', sort=False, seed=0, transforms=[]):
# initialize super class StandardEoDataset
- super().__init__(root_dir, **kwargs)
+ super().__init__(root_dir, use_bands, tile_size, pad, gt_pattern,
+ sort, seed, transforms)
def get_size(self):
+ """Image size of the ProSnow Obergurgl dataset.
+
+ Returns
+ -------
+ size : `tuple`
+ The image size (height, width).
+
+ """
return (310, 270)
class Cloud95Dataset(ImageDataset):
-
- def __init__(self, root_dir, **kwargs):
+ """Dataset class for the `Cloud95`_ dataset by `Mohajerani et al. (2020)`_.
+
+ .. _Cloud95:
+ https://github.com/SorourMo/95-Cloud-An-Extension-to-38-Cloud-Dataset
+ .. _`Mohajerani et al. (2020)`:
+ https://arxiv.org/abs/2001.08768
+
+ Parameters
+ ----------
+ root_dir : `str`
+ The root directory, path to the dataset.
+ use_bands : `list` [`str`], optional
+ A list of the spectral bands to use. The default is [].
+ tile_size : `int` or `None`, optional
+ The size of the tiles. If not `None`, each scene is divided into square
+ tiles of shape (tile_size, tile_size). The default is None.
+ pad : `bool`, optional
+ Whether to center pad the input image. Set ``pad`` = True, if the
+ images are not evenly divisible by the ``tile_size``. The image data is
+ padded with a constant padding value of zero. For each image, the
+ corresponding ground truth image is padded with a "no data" label.
+ The default is False.
+ gt_pattern : `str`, optional
+ A pattern to match the ground truth naming convention. All directories
+ and subdirectories in ``root_dir`` are searched for files matching
+ ``gt_pattern``. The default is '*gt.tif'.
+ sort : `bool`, optional
+ Whether to chronologically sort the samples. Useful for time series
+ data. The default is False.
+ seed : `int`, optional
+ The random seed. Used to split the dataset into training, validation
+ and test set. Useful for reproducibility. The default is 0.
+ transforms : `list` [`pysegcnn.core.split.Augment`], optional
+ List of `pysegcnn.core.split.Augment` instances. Each item in
+ ``transforms`` generates a distinct transformed version of the dataset.
+ The total dataset is composed of the original untransformed dataset
+ together with each transformed version of it.
+ If ``transforms`` = [], only the original dataset is used.
+ The default is [].
+
+ Returns
+ -------
+ None.
+
+ """
+
+ def __init__(self, root_dir, use_bands=[], tile_size=None, pad=False,
+ gt_pattern='*gt.tif', sort=False, seed=0, transforms=[]):
+ # initialize super class StandardEoDataset
+ super().__init__(root_dir, use_bands, tile_size, pad, gt_pattern,
+ sort, seed, transforms)
# the csv file containing the names of the informative patches
# patches with more than 80% black pixels, i.e. patches resulting from
# the black margins around a Landsat 8 scene are excluded
self.exclude = 'training_patches_95-cloud_nonempty.csv'
- # initialize super class ImageDataset
- super().__init__(root_dir, **kwargs)
-
- # image size of the Cloud-95 dataset: (height, width)
def get_size(self):
+ """Image size of the Cloud-95 dataset.
+
+ Returns
+ -------
+ size : `tuple`
+ The image size (height, width).
+
+ """
return (384, 384)
- # Landsat 8 bands in the Cloud-95 dataset
def get_sensor(self):
+ """Landsat 8 bands of the Cloud-95 dataset.
+
+ Returns
+ -------
+ sensor : `enum.Enum`
+ An enumeration of the bands of the sensor.
+
+ """
return Landsat8
- # class labels of the Cloud-95 dataset
def get_labels(self):
+ """Class labels of the Cloud-95 dataset.
+
+ Returns
+ -------
+ labels : `enum.Enum`
+ The class labels.
+
+ """
return Cloud95Labels
- # preprocess Cloud-95 dataset
def preprocess(self, data, gt):
-
+ """Preprocess Cloud-95 dataset images.
+
+ Parameters
+ ----------
+ data : `numpy.ndarray`
+ The sample input data.
+ gt : `numpy.ndarray`
+ The sample ground truth.
+
+ Returns
+ -------
+ data : `numpy.ndarray`
+ The preprocessed input data.
+ gt : `numpy.ndarray`
+ The preprocessed ground truth data.
+
+ """
# normalize the data
# here, we use the normalization of the authors of Cloud-95, i.e.
# Mohajerani and Saeedi (2019, 2020)
data /= 65535
gt[gt != self.cval] /= 255
-
return data, gt
- # function that parses the date from a Landsat 8 scene id
- def parse_scene_id(self, scene):
- return parse_landsat_scene(scene)
+ def parse_scene_id(self, scene_id):
+ """Parse Sparcs scene identifiers (Landsat 8).
- def compose_scenes(self):
+ Parameters
+ ----------
+ scene_id : `str`
+ A scene identifier.
+
+ Returns
+ -------
+ scene : `dict` or `None`
+ A dictionary containing scene metadata. If `None`, ``scene_id`` is
+ not a valid Landsat scene identifier.
+ """
+ return parse_landsat_scene(scene_id)
+
+ def compose_scenes(self):
+ """Build the list of samples of the dataset.
+
+ Each sample is represented by a dictionary.
+
+ Returns
+ -------
+ scenes : `list` [`dict`]
+ Each item in ``scenes`` is a `dict` with keys:
+ ``'band_name_1'``
+ Path to the file of band_1.
+ ``'band_name_2'``
+ Path to the file of band_2.
+ ``'band_name_n'``
+ Path to the file of band_n.
+ ``'gt'``
+ Path to the ground truth file.
+ ``'date'``
+ The date of the sample.
+ ``'tile'``
+ The tile id of the sample.
+ ``'transform'``
+ The transformation to apply.
+ ``'id'``
+ The scene identifier.
+
+ """
# whether to exclude patches with more than 80% black pixels
ipatches = []
if self.exclude is not None:
@@ -620,7 +1341,7 @@ class Cloud95Dataset(ImageDataset):
patchname = file.split('.')[0].replace(biter + '_', '')
# get the date of the current scene
- date = self.parse_scene_id(patchname)['date']
+ scene_meta = self.parse_scene_id(patchname)
# check whether the current file is an informative patch
if ipatches and patchname not in ipatches:
@@ -641,11 +1362,14 @@ class Cloud95Dataset(ImageDataset):
scene[band] = os.path.join(band_dirs[band],
file.replace(biter, band))
+ # the name of the scene the patch was extracted from
+ scene['id'] = scene_meta['id']
+
# store tile number
scene['tile'] = tile
# store date
- scene['date'] = date
+ scene['date'] = scene_meta['date']
# store optional transformation
scene['transform'] = transf
@@ -661,49 +1385,9 @@ class Cloud95Dataset(ImageDataset):
class SupportedDatasets(enum.Enum):
+ """Names and corresponding classes of the implemented datasets."""
+
Sparcs = SparcsDataset
Cloud95 = Cloud95Dataset
Garmisch = ProSnowGarmisch
Obergurgl = ProSnowObergurgl
-
-
-if __name__ == '__main__':
-
- # define path to working directory
- # wd = '//projectdata.eurac.edu/projects/cci_snow/dfrisinghelli/'
- # wd = '/mnt/CEPH_PROJECTS/cci_snow/dfrisinghelli'
- wd = 'C:/Eurac/2020/'
-
- # path to the preprocessed sparcs dataset
- sparcs_path = os.path.join(wd, '_Datasets/Sparcs')
-
- # path to the Cloud-95 dataset
- # cloud_path = os.path.join(wd, '_Datasets/Cloud95/Training')
-
- # path to the ProSnow dataset
- # prosnow_path = os.path.join(wd, '_Datasets/ProSnow/')
-
- # instanciate the Cloud-95 dataset
- # cloud_dataset = Cloud95Dataset(cloud_path,
- # tile_size=192,
- # use_bands=[],
- # sort=False)
-
- # instanciate the SparcsDataset class
- sparcs_dataset = SparcsDataset(sparcs_path,
- tile_size=None,
- use_bands=['red', 'green', 'blue', 'nir'],
- sort=False,
- transforms=[],
- gt_pattern='*mask.png',
- pad=True,
- cval=99,
- )
-
- # instanciate the ProSnow datasets
- # garmisch = ProSnowGarmisch(os.path.join(prosnow_path, 'Garmisch'),
- # tile_size=None,
- # use_bands=['nir', 'red', 'green', 'blue'],
- # sort=True,
- # transforms=[],
- # gt_pattern='*class.img')
diff --git a/pysegcnn/core/graphics.py b/pysegcnn/core/graphics.py
index 5d765cdfe2d08ec2b3aeb6048765ddd87d1e35ad..e760d9dda5c3b995bd406db3e0bf5424476a703f 100644
--- a/pysegcnn/core/graphics.py
+++ b/pysegcnn/core/graphics.py
@@ -83,15 +83,15 @@ def plot_sample(x, y, use_bands, labels, y_pred=None, figsize=(10, 10),
Parameters
----------
- x : `numpy.ndarray` or `torch.tensor`, (b, h, w)
+ x : `numpy.ndarray` or `torch.Tensor`, (b, h, w)
Array containing the raw data of the tile, shape=(bands, height, width)
- y : `numpy.ndarray` or `torch.tensor`, (h, w)
+ y : `numpy.ndarray` or `torch.Tensor`, (h, w)
Array containing the ground truth of tile ``x``, shape=(height, width)
use_bands : `list` of `str`
List describing the order of the bands in ``x``.
labels : `dict` [`int`, `dict`]
- The keys are the values of the class labels in the ground truth ``y``.
- Each nested `dict` should have keys:
+ The label dictionary. The keys are the values of the class labels
+ in the ground truth ``y``. Each nested `dict` should have keys:
``'color'``
A named color (`str`).
``'label'``
@@ -185,8 +185,8 @@ def plot_confusion_matrix(cm, labels, normalize=True,
cm : `numpy.ndarray`
The confusion matrix.
labels : `dict` [`int`, `dict`]
- The keys are the values of the class labels in the ground truth ``y``.
- Each nested `dict` should have keys:
+ The label dictionary. The keys are the values of the class labels
+ in the ground truth ``y``. Each nested `dict` should have keys:
``'color'``
A named color (`str`).
``'label'``
diff --git a/pysegcnn/core/layers.py b/pysegcnn/core/layers.py
index 367bd7259537e08bf19cad04ebeb9a42a36c164d..ab6c59a1b85b5c1f9065a54c807b315fb2b60633 100644
--- a/pysegcnn/core/layers.py
+++ b/pysegcnn/core/layers.py
@@ -134,16 +134,16 @@ class Conv2dPool(nn.Module):
Parameters
----------
- x : `torch.tensor`
+ x : `torch.Tensor`
Output of previous layer.
Returns
-------
- y : `torch.tensor`
+ y : `torch.Tensor`
Output of this block.
- x : `torch.tensor`
+ x : `torch.Tensor`
Output before max pooling. Stored for skip connections.
- i : `torch.tensor`
+ i : `torch.Tensor`
Indices of the max pooling operation. Used in unpooling operation.
"""
@@ -190,18 +190,18 @@ class Conv2dUnpool(nn.Module):
Parameters
----------
- x : `torch.tensor`
+ x : `torch.Tensor`
Output of previous layer.
- feature : `torch.tensor`
+ feature : `torch.Tensor`
Encoder feature used for the skip connection.
- indices : `torch.tensor`
+ indices : `torch.Tensor`
Indices of the max pooling operation. Used in unpooling operation.
skip : `bool`
Whether to apply skip connetion.
Returns
-------
- x : `torch.tensor`
+ x : `torch.Tensor`
Output of this block.
"""
@@ -254,11 +254,11 @@ class Conv2dUpsample(nn.Module):
Parameters
----------
- x : `torch.tensor`
+ x : `torch.Tensor`
Output of previous layer.
- feature : `torch.tensor`
+ feature : `torch.Tensor`
Encoder feature used for the skip connection.
- indices : `torch.tensor`
+ indices : `torch.Tensor`
Indices of the max pooling operation. Used in unpooling operation.
Not used here, but passed to preserve generic interface. Useful in
`pysegcnn.core.layers.Decoder`.
@@ -267,7 +267,7 @@ class Conv2dUpsample(nn.Module):
Returns
-------
- x : `torch.tensor`
+ x : `torch.Tensor`
Output of this block.
"""
@@ -331,12 +331,12 @@ class Encoder(nn.Module):
Parameters
----------
- x : `torch.tensor`
+ x : `torch.Tensor`
Input image.
Returns
-------
- x : `torch.tensor`
+ x : `torch.Tensor`
Output of the encoder.
"""
@@ -411,7 +411,7 @@ class Decoder(nn.Module):
Parameters
----------
- x : `torch.tensor`
+ x : `torch.Tensor`
Output of the encoder.
enc_cache : `dict`
Cache dictionary with keys:
@@ -422,7 +422,7 @@ class Decoder(nn.Module):
Returns
-------
- x : `torch.tensor`
+ x : `torch.Tensor`
Output of the decoder.
"""
diff --git a/pysegcnn/core/models.py b/pysegcnn/core/models.py
index 52400d7fb361cc2d01127f4d08d781c889e9fc92..0cd85f4db17a8bf3938d4865280d0674be1ebaeb 100644
--- a/pysegcnn/core/models.py
+++ b/pysegcnn/core/models.py
@@ -275,7 +275,7 @@ class UNet(Network):
Parameters
----------
- x : `torch.tensor`
+ x : `torch.Tensor`
The input image, shape=(batch_size, channels, height, width).
Returns
diff --git a/pysegcnn/core/split.py b/pysegcnn/core/split.py
index ea6bef9e24f31c65a3c04765f705a9837edb7ce9..2be0aacbc7e9d56cdb5674b6651de16054a0ad71 100644
--- a/pysegcnn/core/split.py
+++ b/pysegcnn/core/split.py
@@ -214,9 +214,9 @@ def date_scene_split(ds, date, dateformat='%Y%m%d'):
----------
ds : `pysegcnn.core.dataset.ImageDataset`
An instance of `~pysegcnn.core.dataset.ImageDataset`.
- date : 'str'
+ date : `str`
A date.
- dateformat : 'str', optional
+ dateformat : `str`, optional
The format of ``date``. ``dateformat`` is used by
`datetime.datetime.strptime' to parse ``date`` to a `datetime.datetime`
object. The default is '%Y%m%d'.
@@ -283,7 +283,7 @@ def pairwise_disjoint(sets):
class CustomSubset(Subset):
- """Custom subset inheriting `torch.utils.data.dataset.Subset`."""
+ """Custom subset inheriting `torch.utils.data.Subset`."""
def __repr__(self):
"""Representation of ``~pysegcnn.core.split.CustomSubset``."""
diff --git a/pysegcnn/core/trainer.py b/pysegcnn/core/trainer.py
index 31c615f3156d7832917389c53059ce5f8e8bc315..1e0916147652bae3c52e4ec59f33c35a81826366 100644
--- a/pysegcnn/core/trainer.py
+++ b/pysegcnn/core/trainer.py
@@ -1,9 +1,8 @@
+"""Model configuration and training."""
+
+# !/usr/bin/env python
# -*- coding: utf-8 -*-
-"""
-Created on Wed Aug 12 10:24:34 2020
-@author: Daniel
-"""
# builtins
import dataclasses
import pathlib
@@ -34,8 +33,22 @@ LOGGER = logging.getLogger(__name__)
@dataclasses.dataclass
class BaseConfig:
+ """Base `dataclasses.dataclass` for each configuration."""
def __post_init__(self):
+ """Check the type of each argument.
+
+ Raises
+ ------
+ TypeError
+ Raised if the conversion to the specified type of the argument
+ fails.
+
+ Returns
+ -------
+ None.
+
+ """
# check input types
for field in dataclasses.fields(self):
# the value of the current field
@@ -55,6 +68,45 @@ class BaseConfig:
@dataclasses.dataclass
class DatasetConfig(BaseConfig):
+ """Dataset configuration class.
+
+ Parameters
+ ----------
+ dataset_name : `str`
+ The name of the dataset.
+ root_dir : `pathlib.Path`
+ The root directory, path to the dataset.
+ bands : `list` [`str`]
+ A list of the spectral bands to use.
+ tile_size : `int`
+ The size of the tiles. Each scene is divided into square tiles of shape
+ (tile_size, tile_size).
+ gt_pattern : `str`
+ A pattern to match the ground truth naming convention. All directories
+ and subdirectories in ``root_dir`` are searched for files matching
+ ``gt_pattern``.
+ seed : `int`
+ The random seed. Used to split the dataset into training, validation
+ and test set. Useful for reproducibility. The default is 0.
+ sort : `bool`, optional
+ Whether to chronologically sort the samples. Useful for time series
+ data. The default is False.
+ transforms : `list` [`pysegcnn.core.split.Augment`], optional
+ List of `pysegcnn.core.split.Augment` instances. Each item in
+ ``transforms`` generates a distinct transformed version of the dataset.
+ The total dataset is composed of the original untransformed dataset
+ together with each transformed version of it.
+ If ``transforms`` = [], only the original dataset is used.
+ The default is [].
+ pad : `bool`, optional
+ Whether to center pad the input image. Set ``pad`` = True, if the
+ images are not evenly divisible by the ``tile_size``. The image data is
+ padded with a constant padding value of zero. For each image, the
+ corresponding ground truth image is padded with a "no data" label.
+ The default is False.
+
+ """
+
dataset_name: str
root_dir: pathlib.Path
bands: list
@@ -66,6 +118,21 @@ class DatasetConfig(BaseConfig):
pad: bool = False
def __post_init__(self):
+ """Check the type of each argument.
+
+ Raises
+ ------
+ FileNotFoundError
+ Raised if ``root_dir`` does not exist.
+ TypeError
+ Raised if not each item in ``transforms`` is an instance of
+ `pysegcnn.core.split.Augment` in case ``transforms`` is not empty.
+
+ Returns
+ -------
+ None.
+
+ """
# check input types
super().__post_init__()
@@ -84,7 +151,14 @@ class DatasetConfig(BaseConfig):
Augment.__name__])))
def init_dataset(self):
+ """Instanciate the dataset.
+
+ Returns
+ -------
+ dataset : `pysegcnn.core.dataset.ImageDataset`
+ An instance of `pysegcnn.core.dataset.ImageDataset`.
+ """
# instanciate the dataset
dataset = self.dataset_class(
root_dir=str(self.root_dir),
@@ -102,6 +176,32 @@ class DatasetConfig(BaseConfig):
@dataclasses.dataclass
class SplitConfig(BaseConfig):
+ """Dataset split configuration class.
+
+ Parameters
+ ----------
+ split_mode : `str`
+ The mode to split the dataset.
+ ttratio : `float`
+ The ratio of training and validation data to test data, e.g.
+ ``ttratio`` = 0.6 means 60% for training and validation, 40% for
+ testing.
+ tvratio : `float`
+ The ratio of training data to validation data, e.g. ``tvratio`` = 0.8
+ means 80% training, 20% validation.
+ date : `str`, optional
+ A date. Used if ``split_mode`` = 'date'. The default is 'yyyymmdd'.
+ dateformat : `str`, optional
+ The format of ``date``. ``dateformat`` is used by
+ `datetime.datetime.strptime' to parse ``date`` to a `datetime.datetime`
+ object. The default is '%Y%m%d'.
+ drop : `float`, optional
+ Whether to drop samples (during training only) with a fraction of
+ pixels equal to the constant padding value >= ``drop``. ``drop`` = 0
+ means, do not drop any samples. The default is 0.
+
+ """
+
split_mode: str
ttratio: float
tvratio: float
@@ -110,17 +210,46 @@ class SplitConfig(BaseConfig):
drop: float = 0
def __post_init__(self):
+ """Check the type of each argument.
+
+ Raises
+ ------
+ ValueError
+ Raised if ``split_mode`` is not supported.
+
+ Returns
+ -------
+ None.
+
+ """
# check input types
super().__post_init__()
# check if the split mode is valid
self.split_class = item_in_enum(self.split_mode, SupportedSplits)
- # function to drop samples with a fraction of pixels equal to the constant
- # padding value self.cval >= self.drop
@staticmethod
def _drop_samples(ds, drop_threshold=1):
-
+ """Drop samples with a fraction of pixels equal to the padding value.
+
+ Parameters
+ ----------
+ ds : `pysegcnn.core.split.RandomSubset` or
+ `pysegcnn.core.split.SceneSubset`.
+ An instance of `pysegcnn.core.split.RandomSubset` or
+ `pysegcnn.core.split.SceneSubset`.
+ drop_threshold : `float`, optional
+ The threshold above which samples are dropped. ``drop_threshold`` =
+ 1 means a sample is dropped, if all pixels are equal to the padding
+ value. ``drop_threshold`` = 0.8 means, drop a sample if 80% of the
+ pixels are equal to the padding value, etc. The default is 1.
+
+ Returns
+ -------
+ dropped : `list` [`dict`]
+ List of the dropped samples.
+
+ """
# iterate over the scenes returned by self.compose_scenes()
dropped = []
for pos, i in enumerate(ds.indices):
@@ -145,7 +274,32 @@ class SplitConfig(BaseConfig):
return dropped
def train_val_test_split(self, ds):
-
+ """Split ``ds`` into training, validation and test set.
+
+ Parameters
+ ----------
+ ds : `pysegcnn.core.dataset.ImageDataset`
+ An instance of `pysegcnn.core.dataset.ImageDataset`.
+
+ Raises
+ ------
+ TypeError
+ Raised if ``ds`` is not an instance of
+ `pysegcnn.core.dataset.ImageDataset`.
+
+ Returns
+ -------
+ train_ds : `pysegcnn.core.split.RandomSubset` or
+ `pysegcnn.core.split.SceneSubset`.
+ The training set.
+ valid_ds : `pysegcnn.core.split.RandomSubset` or
+ `pysegcnn.core.split.SceneSubset`.
+ The validation set.
+ test_ds : `pysegcnn.core.split.RandomSubset` or
+ `pysegcnn.core.split.SceneSubset`.
+ The test set.
+
+ """
if not isinstance(ds, ImageDataset):
raise TypeError('Expected "ds" to be {}.'
.format('.'.join([ImageDataset.__module__,
@@ -172,6 +326,31 @@ class SplitConfig(BaseConfig):
@staticmethod
def dataloaders(*args, **kwargs):
+ """Build `torch.utils.data.DataLoader` instances.
+
+ Parameters
+ ----------
+ *args : `list` [`torch.utils.data.Dataset`]
+ List of instances of `torch.utils.data.Dataset`.
+ **kwargs
+ Additional keyword arguments passed to
+ `torch.utils.data.DataLoader`.
+
+ Raises
+ ------
+ TypeError
+ Raised if not each item in ``args`` is an instance of
+ `torch.utils.data.Dataset`.
+
+ Returns
+ -------
+ loaders : `list` [`torch.utils.data.DataLoader`]
+ List of instances of `torch.utils.data.DataLoader`. If an instance
+ of `torch.utils.data.Dataset` in ``args`` is empty, `None` is
+ appended to ``loaders`` instead of an instance of
+ `torch.utils.data.DataLoader`.
+
+ """
# check whether each dataset in args has the correct type
loaders = []
for ds in args:
@@ -192,6 +371,72 @@ class SplitConfig(BaseConfig):
@dataclasses.dataclass
class ModelConfig(BaseConfig):
+ """Model configuration class.
+
+ Parameters
+ ----------
+ model_name : `str`
+ The name of the model.
+ filters : `list` [`int`]
+ List of input channels to the convolutional layers.
+ torch_seed : `int`
+ The random seed to initialize the model weights.
+ Useful for reproducibility.
+ optim_name : `str`
+ The name of the optimizer to update the model weights.
+ loss_name : `str`
+ The name of the loss function measuring the model error.
+ skip_connection : `bool`, optional
+ Whether to apply skip connections. The defaul is True.
+ kwargs: `dict`, optional
+ The configuration for each convolution in the model. The default is
+ {'kernel_size': 3, 'stride': 1, 'dilation': 1}.
+ batch_size : `int`, optional
+ The model batch size. Determines the number of samples to process
+ before updating the model weights. The default is 64.
+ checkpoint : `bool`, optional
+ Whether to resume training from an existing model checkpoint. The
+ default is False.
+ transfer : `bool`, optional
+ Whether to use a model for transfer learning on a new dataset. If True,
+ the model architecture of ``pretrained_model`` is adjusted to a new
+ dataset. The default is False.
+ pretrained_model : `str`, optional
+ The name of the pretrained model to use for transfer learning.
+ The default is ''.
+ lr : `float`, optional
+ The learning rate used by the gradient descent algorithm.
+ The default is 0.001.
+ early_stop : `bool`, optional
+ Whether to apply `early stopping`_. The default is False.
+ mode : `str`, optional
+ The mode of the early stopping. Depends on the metric measuring
+ performance. When using model loss as metric, use ``mode`` = 'min',
+ however, when using accuracy as metric, use ``mode`` = 'max'. For now,
+ only ``mode`` = 'max' is supported. Only used if ``early_stop`` = True.
+ The default is 'max'.
+ delta : `float`, optional
+ Minimum change in early stopping metric to be considered as an
+ improvement. Only used if ``early_stop`` = True. The default is 0.
+ patience : `int`, optional
+ The number of epochs to wait for an improvement in the early stopping
+ metric. If the model does not improve over more than ``patience``
+ epochs, quit training. Only used if ``early_stop`` = True.
+ The default is 10.
+ epochs : `int`, optional
+ The maximum number of epochs to train. The default is 50.
+ nthreads : `int`, optional
+ The number of cpu threads to use during training. The default is
+ torch.get_num_threads().
+ save : `bool`, optional
+ Whether to save the model state to disk. Model states are saved in
+ pysegcnn/main/_models. The default is True.
+
+ .. _early stopping:
+ https://en.wikipedia.org/wiki/Early_stopping
+
+ """
+
model_name: str
filters: list
torch_seed: int
@@ -214,6 +459,19 @@ class ModelConfig(BaseConfig):
save: bool = True
def __post_init__(self):
+ """Check the type of each argument.
+
+ Raises
+ ------
+ ValueError
+ Raised if the model ``model_name``, the optimizer ``optim_name`` or
+ the loss function ``loss_name`` is not supported.
+
+ Returns
+ -------
+ None.
+
+ """
# check input types
super().__post_init__()
@@ -233,7 +491,19 @@ class ModelConfig(BaseConfig):
self.pretrained_path = self.state_path.joinpath(self.pretrained_model)
def init_optimizer(self, model):
+ """Instanciate the optimizer.
+ Parameters
+ ----------
+ model : `torch.nn.Module`
+ An instance of `torch.nn.Module`.
+
+ Returns
+ -------
+ optimizer : `torch.optim.Optimizer`
+ An instance of `torch.optim.Optimizer`.
+
+ """
LOGGER.info('Optimizer: {}.'.format(repr(self.optim_class)))
# initialize the optimizer for the specified model
@@ -242,7 +512,14 @@ class ModelConfig(BaseConfig):
return optimizer
def init_loss_function(self):
+ """Instanciate the loss function.
+
+ Returns
+ -------
+ loss_function : `torch.nn.Module`
+ An instance of `torch.nn.Module`.
+ """
LOGGER.info('Loss function: {}.'.format(repr(self.loss_class)))
# instanciate the loss function
@@ -251,7 +528,38 @@ class ModelConfig(BaseConfig):
return loss_function
def init_model(self, ds, state_file):
-
+ """Instanciate the model and the optimizer.
+
+ If the model checkpoint ``state_file`` exists, the pretrained model and
+ optimizer states are loaded, otherwise the model and the optimizer are
+ initialized from scratch.
+
+ Parameters
+ ----------
+ ds : `pysegcnn.core.dataset.ImageDataset`
+ An instance of `pysegcnn.core.dataset.ImageDataset`.
+ state_file : `pathlib.Path`
+ Path to a model checkpoint.
+
+ Returns
+ -------
+ model : `pysegcnn.core.models.Network`
+ An instance of `pysegcnn.core.models.Network`.
+ optimizer : `torch.optim.Optimizer`
+ An instance of `torch.optim.Optimizer`.
+ checkpoint_state : `dict`
+ If the model checkpoint ``state_file`` exists, ``checkpoint_state``
+ has keys:
+ ``'ta'``
+ The accuracy on the training set (`numpy.ndarray`).
+ ``'tl'``
+ The loss on the training set (`numpy.ndarray`).
+ ``'va'``
+ The accuracy on the validation set (`numpy.ndarray`).
+ ``'vl'``
+ The loss on the validation set (`numpy.ndarray`).
+
+ """
# write an initialization string to the log file
LogConfig.init_log('{}: Initializing model run. ')
@@ -290,7 +598,39 @@ class ModelConfig(BaseConfig):
@staticmethod
def load_checkpoint(model, optimizer, state_file):
-
+ """Load an existing model checkpoint.
+
+ If the model checkpoint ``state_file`` exists, the pretrained model and
+ optimizer states are loaded.
+
+ Parameters
+ ----------
+ model : `pysegcnn.core.models.Network`
+ An instance of `pysegcnn.core.models.Network`.
+ optimizer : `torch.optim.Optimizer`
+ An instance of `torch.optim.Optimizer`.
+ state_file : `pathlib.Path`
+ Path to the model checkpoint.
+
+ Returns
+ -------
+ model : `pysegcnn.core.models.Network`
+ An instance of `pysegcnn.core.models.Network`.
+ optimizer : `torch.optim.Optimizer`
+ An instance of `torch.optim.Optimizer`.
+ checkpoint_state : `dict`
+ If the model checkpoint ``state_file`` exists, ``checkpoint_state``
+ has keys:
+ ``'ta'``
+ The accuracy on the training set (`numpy.ndarray`).
+ ``'tl'``
+ The loss on the training set (`numpy.ndarray`).
+ ``'va'``
+ The accuracy on the validation set (`numpy.ndarray`).
+ ``'vl'``
+ The loss on the validation set (`numpy.ndarray`).
+
+ """
# whether to resume training from an existing model checkpoint
checkpoint_state = {}
@@ -315,7 +655,36 @@ class ModelConfig(BaseConfig):
@staticmethod
def transfer_model(state_file, ds):
-
+ """Adjust a pretrained model to a new dataset.
+
+ The classification layer of the pretrained model in ``state_file`` is
+ initilialized from scratch with the classes of the new dataset ``ds``.
+
+ The remaining model weights are preserved.
+
+ Parameters
+ ----------
+ state_file : `pathlib.Path`
+ Path to a pretrained model.
+ ds : `pysegcnn.core.dataset.ImageDataset`
+ An instance of `pysegcnn.core.dataset.ImageDataset`.
+
+ Raises
+ ------
+ TypeError
+ Raised if ``ds`` is not an instance of
+ `pysegcnn.core.dataset.ImageDataset`.
+ ValueError
+ Raised if the bands of ``ds`` do not match the bands of the dataset
+ the pretrained model was trained with.
+
+ Returns
+ -------
+ model : `pysegcnn.core.models.Network`
+ An instance of `pysegcnn.core.models.Network`. The pretrained model
+ adjusted to the new dataset.
+
+ """
# check input type
if not isinstance(ds, ImageDataset):
raise TypeError('Expected "ds" to be {}.'
@@ -346,8 +715,7 @@ class ModelConfig(BaseConfig):
.format(', '.join('({}, {})'.format(k, v['label'])
for k, v in ds.labels.items())))
- # adjust the classification layer to the number of classes of the
- # current dataset
+ # adjust the classification layer to the classes of the new dataset
model.classifier = Conv2dSame(in_channels=filters[0],
out_channels=model.nclasses,
kernel_size=1)
@@ -757,7 +1125,7 @@ class EarlyStopping(object):
# whether to check for an increase or a decrease in a given metric
self.is_better = self.decreased if mode == 'min' else self.increased
- # minimum change in metric to be classified as an improvement
+ # minimum change in metric to be considered as an improvement
self.min_delta = min_delta
# number of epochs to wait for improvement