diff --git a/docs/conf.py b/docs/conf.py
index c3ffaec690db1aad2cec9bdc03662c2f87fa6b5f..c4ba79164c898cd0cdb8f31610b20c1fca61ba89 100644
--- a/docs/conf.py
+++ b/docs/conf.py
@@ -34,15 +34,23 @@ extensions = [
'sphinx.ext.autodoc',
'sphinx.ext.autosummary',
'sphinx.ext.viewcode',
+ 'sphinx.ext.intersphinx',
'numpydoc',
]
# numpydoc options
numpydoc_show_class_members = False
numpydoc_show_inherited_class_members = False
+numpydoc_attributes_as_param_list = True
+# numpydoc_xref_param_type = True
+# numpydoc_xref_ignore = {'type', 'optional', 'default'}
# autosummary options
autosummary_generate = True
+autodoc_default_options = {
+ 'inherited-members': False,
+ 'show-inheritance': True,
+ }
# Add any paths that contain templates here, relative to this directory.
templates_path = ['_templates']
diff --git a/pysegcnn/core/cli.py b/pysegcnn/core/cli.py
index 9576209e62a85c9eeb966d10ad32831eaf54056d..07c8759bdb50db3af6cfedebc4c29eeed6c0122e 100644
--- a/pysegcnn/core/cli.py
+++ b/pysegcnn/core/cli.py
@@ -18,19 +18,12 @@ License
import pathlib
import argparse
-
# epilogue to display at the end of each parser
EPILOGUE = 'Author: Daniel Frisinghelli, daniel.frisinghelli@gmail.com'
def structure_parser():
- """Command line argument parser to standardize dataset structure.
-
- Returns
- -------
- None.
-
- """
+ """Command line argument parser to standardize dataset structure."""
parser = argparse.ArgumentParser(
description='Standardize the dataset directory structure.',
epilog=EPILOGUE,
diff --git a/pysegcnn/core/constants.py b/pysegcnn/core/constants.py
index ffe0e82e33f73e83e23e9745f40ee9cb3b16413c..491efe1b5822bbbc7c9be840ec8d1fd3611f9d9b 100644
--- a/pysegcnn/core/constants.py
+++ b/pysegcnn/core/constants.py
@@ -1,4 +1,4 @@
-"""A collection of constant values.
+"""A collection of constants.
License
-------
@@ -20,12 +20,15 @@ import enum
# Landsat 8 bands
class Landsat8(enum.Enum):
- """The spectral bands of the Landsat 8 sensors.
+ """The spectral bands of the `Landsat-8`_ sensors.
sensors:
- Operational Land Imager (OLI), (bands 1-9)
- Thermal Infrared Sensor (TIRS), (bands 10, 11)
+ .. _Landsat-8:
+ https://www.usgs.gov/land-resources/nli/landsat/landsat-8?qt-science_support_page_related_con=0#qt-science_support_page_related_con
+
"""
violet = 1
@@ -43,7 +46,12 @@ class Landsat8(enum.Enum):
# Sentinel 2 bands
class Sentinel2(enum.Enum):
- """The spectral bands of the Sentinel-2 MultiSpectral Instrument (MSI)."""
+ """The spectral bands of the `Sentinel-2`_ MultiSpectral Instrument (MSI).
+
+ .. _Sentinel-2:
+ https://sentinel.esa.int/web/sentinel/missions/sentinel-2
+
+ """
aerosol = 1
blue = 2
diff --git a/pysegcnn/core/dataset.py b/pysegcnn/core/dataset.py
index d8dfc714e9ed539f7a5b404573e363a603a5943b..b8eee01491ccbdbfd0468a77166227468d8ae844 100644
--- a/pysegcnn/core/dataset.py
+++ b/pysegcnn/core/dataset.py
@@ -45,54 +45,108 @@ LOGGER = logging.getLogger(__name__)
class ImageDataset(Dataset):
- r"""Base class for multispectral image data.
+ """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.
+ Inheriting from :py:class:`torch.utils.data.Dataset` to be compliant to the
+ PyTorch standard. This enables the use of the handy
+ :py:class:`torch.utils.data.DataLoader` class during model training.
- Parameters
+ Attributes
----------
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
+ use_bands : `list` [`str`]
+ List of the spectral bands to use during model training.
+ tile_size : `int` or `None`
+ The size of the tiles.
+ pad : `bool`
+ Whether to center pad the input image.
+ gt_pattern : `str`
A regural expression 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.
+ sort : `bool`
+ Whether to chronologically sort the samples.
+ seed : `int`
+ The random seed.
+ transforms : `list`
+ List of :py:class:`pysegcnn.core.transforms.Augment` instances.
+ size : `tuple` [`int`]
+ The size of an image of the dataset.
+ sensor : :py:class:`enum.Enum`
+ An enumeration of the bands of sensor the dataset is derived from,
+ see e.g. :py:class:`pysegcnn.core.constants.Landsat8`.
+ bands : `dict` [`int`, `str`]
+ The spectral bands of ``sensor``. The keys are the number and the
+ values are the name of the spectral bands.
+ labels : `dict` [`int`, `dict`]
+ The label dictionary. The keys are the values of the class labels
+ in the ground truth. Each nested `dict` has keys:
+ ``'color'``
+ A named color (`str`).
+ ``'label'``
+ The name of the class label (`str`).
+ tiles : `int`
+ Number of tiles with size ``(tile_size, tile_size)`` within an image.
+ padding : `tuple` [`int`]
+ The amount of padding, (bottom, left, top, right).
+ height : `int`
+ The height of a padded image.
+ width : `int`
+ The width of a padded image.
+ topleft : `dict` [`int`, `tuple`]
+ The topleft corners of the tiles. The keys of are the tile ids (`int`)
+ and the values are the topleft corners (y, x) of the tiles.
+ cval : `int`
+ When padding, ``cval`` is the value of the "no data" label in the
+ ground truth. Otherwise, ``cval=0``.
+ gt : `list` [`str` or :py:class:`pathlib.Path`]
+ List of the ground truth images.
+ keys : `list`
+ List of required keys for each dictionary in ``scenes``.
+ scenes : `list` [`dict`]
+ List of dictionaries representing the samples of the dataset.
"""
def __init__(self, root_dir, use_bands=[], tile_size=None, pad=False,
- gt_pattern='(.*)gt.tif', sort=False, seed=0, transforms=[]):
+ gt_pattern='(.*)gt\\.tif', sort=False, seed=0, transforms=[]):
+ r"""Initialize.
+
+ 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 regural expression 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`, optional
+ List of :py:class:`pysegcnn.core.transforms.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 `[]`.
+
+ """
super().__init__()
# dataset configuration
@@ -159,6 +213,9 @@ class ImageDataset(Dataset):
LOGGER.info('Adding label "No data" with value={} to ground truth.'
.format(self.cval))
+ # list of ground truth images
+ self.gt = []
+
def _build_labels(self):
"""Build the label dictionary.
@@ -166,7 +223,7 @@ class ImageDataset(Dataset):
-------
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:
+ in the ground truth. Each nested `dict` should have keys:
``'color'``
A named color (`str`).
``'label'``
@@ -492,7 +549,7 @@ class ImageDataset(Dataset):
-------
stack : `numpy.ndarray`
The input data of the sample.
- gt : TYPE
+ gt : `numpy.ndarray`
The ground truth of the sample.
"""
@@ -503,7 +560,7 @@ class ImageDataset(Dataset):
return stack, gt
def to_tensor(self, x, dtype):
- """Convert ``x`` to `torch.Tensor`.
+ """Convert ``x`` to :py:class:`torch.Tensor`.
Parameters
----------
@@ -565,75 +622,32 @@ class ImageDataset(Dataset):
class StandardEoDataset(ImageDataset):
r"""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:
+ :py:class:`pysegcnn.core.dataset.StandardEoDataset` implements the
+ :py:meth:`~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/
- .
- .
- .
+ - 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.
+ :py:class:`pysegcnn.core.dataset.StandardEoDataset` and implement the
+ remaining methods. If not, you can use
+ :py:func:`pysegcnn.core.utils.standard_eo_structure` to transfer your
+ dataset to the above directory structure.
- 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 regural expression 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.
+ See :py:class:`pysegcnn.core.dataset.SparcsDataset` for an example.
"""
@@ -709,35 +723,9 @@ class StandardEoDataset(ImageDataset):
return scene_data
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.
-
- """
+ """Build the list of samples of the dataset."""
# search the root directory
scenes = []
- self.gt = []
for dirpath, dirname, files in os.walk(self.root):
# search for a ground truth in the current directory
@@ -800,47 +788,13 @@ class StandardEoDataset(ImageDataset):
class SparcsDataset(StandardEoDataset):
- r"""Class for the `Sparcs`_ dataset.
+ """Class for the `Sparcs`_ dataset by `Hughes & Hayes (2014)`_.
.. _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 regural expression 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.
+ .. _Hughes & Hayes (2014):
+ https://www.mdpi.com/2072-4292/6/6/4907
"""
@@ -923,46 +877,7 @@ class SparcsDataset(StandardEoDataset):
class ProSnowDataset(StandardEoDataset):
- r"""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 regural expression 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.
-
- """
+ """Class for the ProSnow datasets."""
def __init__(self, root_dir, use_bands=[], tile_size=None, pad=False,
gt_pattern='(.*)gt\\.tif', sort=False, seed=0, transforms=[]):
@@ -1032,46 +947,7 @@ class ProSnowDataset(StandardEoDataset):
class ProSnowGarmisch(ProSnowDataset):
- r"""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 regural expression 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.
-
- """
+ """Class for the ProSnow Garmisch dataset."""
def __init__(self, root_dir, use_bands=[], tile_size=None, pad=False,
gt_pattern='(.*)gt\\.tif', sort=False, seed=0, transforms=[]):
@@ -1092,46 +968,7 @@ class ProSnowGarmisch(ProSnowDataset):
class ProSnowObergurgl(ProSnowDataset):
- r"""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 regural expression 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.
-
- """
+ """Class for the ProSnow Obergurgl dataset."""
def __init__(self, root_dir, use_bands=[], tile_size=None, pad=False,
gt_pattern='(.*)gt\\.tif', sort=False, seed=0, transforms=[]):
@@ -1152,50 +989,13 @@ class ProSnowObergurgl(ProSnowDataset):
class Cloud95Dataset(ImageDataset):
- r"""Class for the `Cloud-95`_ dataset by `Mohajerani & Saeedi (2020)`_.
+ """Class for the `Cloud-95`_ dataset by `Mohajerani & Saeedi (2020)`_.
.. _Cloud-95:
https://github.com/SorourMo/95-Cloud-An-Extension-to-38-Cloud-Dataset
.. _Mohajerani & Saeedi (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 regural expression 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,
@@ -1285,32 +1085,7 @@ class Cloud95Dataset(ImageDataset):
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.
-
- """
+ """Build the list of samples of the dataset."""
# whether to exclude patches with more than 80% black pixels
ipatches = []
if self.exclude is not None:
diff --git a/pysegcnn/core/layers.py b/pysegcnn/core/layers.py
index b2a51d2c35c7d6bf8cf7876219e07c0f486cca3b..f4fbebab8e642c34ad226978acb17c7fafe62676 100644
--- a/pysegcnn/core/layers.py
+++ b/pysegcnn/core/layers.py
@@ -28,29 +28,30 @@ class Conv2dSame(nn.Conv2d):
calculated such that the output of the convolution has the same spatial
dimensions as the input.
- Parameters
+ Attributes
----------
- *args: `list` [`str`]
- positional arguments passed to `torch.nn.Conv2d`:
- ``'in_channels'``: `int`
- Number of input channels.
- ``'out_channels'``: `int`
- Number of output channels.
- ``'kernel_size'``: `int` or `tuple` [`int`]
- Size of the convolving kernel.
- **kwargs: 'dict' [`str`]
- Additional keyword arguments passed to `torch.nn.Conv2d`_.
-
- .. _torch.nn.Conv2d:
- https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html#torch.nn.Conv2d
-
- Returns
- -------
- None.
+ padding : `tuple` [`int`]
+ The amount of padding, (pad_height, pad_width).
"""
def __init__(self, *args, **kwargs):
+ """Initialize.
+
+ Parameters
+ ----------
+ *args: `list` [`str`]
+ positional arguments passed to :py:class:`torch.nn.Conv2d`:
+ ``'in_channels'``: `int`
+ Number of input channels.
+ ``'out_channels'``: `int`
+ Number of output channels.
+ ``'kernel_size'``: `int` or `tuple` [`int`]
+ Size of the convolving kernel.
+ **kwargs: `dict` [`str`]
+ Additional keyword arguments passed to :py:class:`torch.nn.Conv2d`.
+
+ """
super().__init__(*args, **kwargs)
# define tensorflows "SAME" padding for stride = 1
@@ -73,7 +74,7 @@ class Conv2dSame(nn.Conv2d):
Returns
-------
p : `int`
- the amount of padding.
+ The amount of padding.
"""
# calculates the padding so that the convolution
@@ -90,7 +91,7 @@ def conv_bn_relu(in_channels, out_channels, **kwargs):
Number of input channels.
out_channels : `int`
Number of output channels.
- **kwargs: 'dict' [`str`]
+ **kwargs: `dict` [`str`]
Additional arguments passed to `pysegcnn.core.layers.Conv2dSame`.
Returns
@@ -113,28 +114,39 @@ def conv_bn_relu(in_channels, out_channels, **kwargs):
class Block(nn.Module):
"""Basic convolutional block.
- Parameters
+ Attributes
----------
in_channels : `int`
Number of input channels.
out_channels : `int`
Number of output channels.
- **kwargs: 'dict' [`str`]
- Additional arguments passed to `pysegcnn.core.layers.Conv2dSame`.
-
- Raises
- ------
- TypeError
- Raised if `~pysegcnn.core.layers.Block.layers` method does not return
- an instance of `torch.nn.Sequential`.
-
- Returns
- -------
- None.
+ kwargs: `dict` [`str`]
+ Additional arguments passed to
+ :py:class:`pysegcnn.core.layers.Conv2dSame`.
+ conv : :py:class:`torch.nn.Sequential`
+ The convolutional layers of the block.
"""
def __init__(self, in_channels, out_channels, **kwargs):
+ """Initialize.
+
+ Parameters
+ ----------
+ in_channels : `int`
+ Number of input channels.
+ out_channels : `int`
+ Number of output channels.
+ **kwargs: `dict` [`str`]
+ Additional arguments passed to
+ :py:class:`pysegcnn.core.layers.Conv2dSame`.
+
+ Raises
+ ------
+ TypeError
+ Raised if :py:meth:`~pysegcnn.core.layers.Block.layers` method does
+ not return an instance of :py:class:`torch.nn.Sequential`.
+ """
super().__init__()
# number of input and output channels
@@ -157,13 +169,13 @@ class Block(nn.Module):
Raises
------
NotImplementedError
- Raised if `pysegcnn.core.layers.Block` is not inherited.
+ Raised if :py:class:`pysegcnn.core.layers.Block` is not inherited.
Returns
-------
- layers : `torch.nn.Sequential` [`torch.nn.Module`]
- Return an instance of `torch.nn.Sequential` containing a sequence
- of layer (`torch.nn.Module` ) instances.
+ layers : :py:class:`torch.nn.Sequential` [:py:class:`torch.nn.Module`]
+ Return an instance of :py:class:`torch.nn.Sequential` containing a
+ sequence of layer (:py:class:`torch.nn.Module` ) instances.
"""
raise NotImplementedError('Return an instance of {}.'
@@ -175,33 +187,14 @@ class Block(nn.Module):
Raises
------
NotImplementedError
- Raised if `pysegcnn.core.layers.Block` is not inherited.
-
- Returns
- -------
- None.
+ Raised if :py:class:`pysegcnn.core.layers.Block` is not inherited.
"""
raise NotImplementedError('Implement the forward pass.')
class EncoderBlock(Block):
- """Block of a convolutional encoder.
-
- Parameters
- ----------
- in_channels : `int`
- Number of input channels.
- out_channels : `int`
- Number of output channels.
- **kwargs: 'dict' [`str`]
- Additional arguments passed to `pysegcnn.core.layers.Conv2dSame`.
-
- Returns
- -------
- None.
-
- """
+ """Block of a convolutional encoder."""
def __init__(self, in_channels, out_channels, **kwargs):
super().__init__(in_channels, out_channels, **kwargs)
@@ -211,21 +204,21 @@ class EncoderBlock(Block):
Parameters
----------
- x : `torch.Tensor`
+ x : :py:class:`torch.Tensor`
Input tensor, e.g. output of the previous block/layer.
Returns
-------
- y : `torch.Tensor`, shape=(batch, channel, height, width)
+ y : :py:class:`torch.Tensor`, shape=(b, c, h, w)
Output of the encoder block.
- x : `torch.Tensor`, shape=(batch, channel, height, width)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h, w)
Intermediate output before applying downsampling. Useful to
implement skip connections.
- indices : `torch.Tensor` or `None`
+ indices : :py:class:`torch.Tensor` or `None`
Optional indices of the downsampling method, e.g. indices of the
- maxima when using `torch.nn.functional.max_pool2d`. Useful for
- upsampling later. If no indices are required to upsample, simply
- return ``indices`` = `None`.
+ maxima when using :py:func:`torch.nn.functional.max_pool2d`. Useful
+ for upsampling later. If no indices are required to upsample,
+ simply return ``indices=None``.
"""
# the forward pass of the layers of the block
@@ -239,57 +232,43 @@ class EncoderBlock(Block):
def downsample(self, x):
"""Define the downsampling method.
- The `~pysegcnn.core.layers.EncoderBlock.downsample` method should
- implement the spatial pooling operation.
+ The :py:meth:`~pysegcnn.core.layers.EncoderBlock.downsample` method
+ should implement the spatial pooling operation.
Use one of the following functions to downsample:
- - `torch.nn.functional.max_pool2d`
- - `torch.nn.functional.interpolate`
+ - :py:func:`torch.nn.functional.max_pool2d`
+ - :py:func:`torch.nn.functional.interpolate`
- See `pysegcnn.core.layers.ConvBnReluMaxPool` for an example
+ See :py:class:`pysegcnn.core.layers.ConvBnReluMaxPool` for an example
implementation.
Parameters
----------
- x : `torch.Tensor`, shape=(batch, channel, height, width)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h, w)
Input tensor, e.g. output of a convolutional block.
Raises
------
NotImplementedError
- Raised if `pysegcnn.core.layers.EncoderBlock` is not inherited.
+ Raised if :py:class:`pysegcnn.core.layers.EncoderBlock` is not
+ inherited.
Returns
-------
- x : `torch.Tensor`, shape=(batch, channel, height, width)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h, w)
The spatially downsampled tensor.
- indices : `torch.Tensor` or `None`
+ indices : :py:class:`torch.Tensor` or `None`
Optional indices of the downsampling method, e.g. indices of the
- maxima when using `torch.nn.functional.max_pool2d`. Useful for
- upsampling later. If no indices are required to upsample, simply
- return ``indices`` = `None`.
+ maxima when using :py:func:`torch.nn.functional.max_pool2d`. Useful
+ for upsampling later. If no indices are required to upsample,
+ simply return ``indices=None``.
"""
raise NotImplementedError('Implement the downsampling function.')
class DecoderBlock(Block):
- """Block of a convolutional decoder.
-
- Parameters
- ----------
- in_channels : `int`
- Number of input channels.
- out_channels : `int`
- Number of output channels.
- **kwargs: 'dict' [`str`]
- Additional arguments passed to `pysegcnn.core.layers.Conv2dSame`.
-
- Returns
- -------
- None.
-
- """
+ """Block of a convolutional decoder."""
def __init__(self, in_channels, out_channels, **kwargs):
super().__init__(in_channels, out_channels, **kwargs)
@@ -299,21 +278,21 @@ class DecoderBlock(Block):
Parameters
----------
- x : `torch.Tensor`, shape=(batch, channel, height, width)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h, w)
Input tensor.
- feature : `torch.Tensor`, shape=(batch, channel, height, width)
- Intermediate output of a layer in the encoder.
- If ``skip`` = True, ``feature`` is concatenated (along the channel
- axis) to the output of the respective upsampling layer in the
- decoder (skip connection).
- indices : `torch.Tensor` or `None`
+ feature : :py:class:`torch.Tensor`, shape=(b, c, h, w)
+ Intermediate output of a layer in the encoder. If ``skip=True``,
+ ``feature`` is concatenated (along the channel axis) to the output
+ of the respective upsampling layer in the decoder (skip connection)
+ .
+ indices : :py:class:`torch.Tensor` or `None`
Indices of the encoder downsampling method.
skip : `bool`
Whether to apply the skip connection.
Returns
-------
- x : `torch.Tensor`, shape=(batch, channel, height, width)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h, w)
Output of the decoder block.
"""
@@ -334,35 +313,36 @@ class DecoderBlock(Block):
def upsample(self, x, feature, indices):
"""Define the upsampling method.
- The `~pysegcnn.core.layers.DecoderBlock.upsample` method should
- implement the spatial upsampling operation.
+ The :py:meth:`~pysegcnn.core.layers.DecoderBlock.upsample` method
+ should implement the spatial upsampling operation.
Use one of the following functions to upsample:
- - `torch.nn.functional.max_unpool2d`
- - `torch.nn.functional.interpolate`
+ - :py:func:`torch.nn.functional.max_unpool2d`
+ - :py:func:`torch.nn.functional.interpolate`
- See `pysegcnn.core.layers.ConvBnReluMaxUnpool` or
- `pysegcnn.core.layers.ConvBnReluUpsample` for an example
+ See :py:class:`pysegcnn.core.layers.ConvBnReluMaxUnpool` or
+ :py:class:`pysegcnn.core.layers.ConvBnReluUpsample` for an example
implementation.
Parameters
----------
- x : `torch.Tensor`, shape=(batch, channel, height, width)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h, w)
Input tensor, e.g. output of a convolutional block.
- feature : `torch.Tensor`, shape=(batch, channel, height, width)
+ feature : :py:class:`torch.Tensor`, shape=(b, c, h, w)
Intermediate output of a layer in the encoder. Used to implement
skip connections.
- indices : `torch.Tensor` or `None`
+ indices : :py:class:`torch.Tensor` or `None`
Indices of the encoder downsampling method.
Raises
------
NotImplementedError
- Raised if `pysegcnn.core.layers.DecoderBlock` is not inherited.
+ Raised if :py:class:`pysegcnn.core.layers.DecoderBlock` is not
+ inherited.
Returns
-------
- x : `torch.Tensor`, shape=(batch, channel, height, width)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h, w)
The spatially upsampled tensor.
"""
@@ -373,31 +353,50 @@ class Encoder(nn.Module):
"""Generic convolutional encoder.
When instanciating an encoder-decoder architechure, ``filters`` should be
- the same for `pysegcnn.core.layers.Encoder` and
- `pysegcnn.core.layers.Decoder`.
+ the same for :py:class:`pysegcnn.core.layers.Encoder` and
+ :py:class:`pysegcnn.core.layers.Decoder`.
- See `pysegcnn.core.models.UNet` for an example implementation.
+ See :py:class:`pysegcnn.core.models.UNet` for an example implementation.
- Parameters
+ Attributes
----------
- filters : `list` [`int`]
- List of input channels to each convolutional block. The length of
- ``filters`` determines the depth of the encoder. The first element of
- ``filters`` has to be the number of channels of the input images.
- block : `pysegcnn.core.layers.EncoderBlock`
+ features : :py:class:`numpy.ndarray`
+ Input channels to each convolutional block, i.e. ``filters``.
+ block : :py:class:`pysegcnn.core.layers.EncoderBlock`
The convolutional block defining a layer in the encoder.
- A subclass of `pysegcnn.core.layers.EncoderBlock`, e.g.
- `pysegcnn.core.layers.ConvBnReluMaxPool`.
- **kwargs: 'dict' [`str`]
- Additional arguments passed to `pysegcnn.core.layers.Conv2dSame`.
-
- Returns
- -------
- None.
+ A subclass of :py:class:`pysegcnn.core.layers.EncoderBlock`, e.g.
+ :py:class:`pysegcnn.core.layers.ConvBnReluMaxPool`.
+ layers : :py:class:`torch.nn.ModuleList`
+ List of blocks in the encoder.
+ cache : `dict`
+ Intermediate encoder outputs. Dictionary with keys:
+ ``'feature'``
+ The intermediate encoder outputs (:py:class:`torch.Tensor`).
+ ``'indices'``
+ The indices of the max pooling layer, if required
+ (:py:class:`torch.Tensor`).
"""
def __init__(self, filters, block, **kwargs):
+ """Initialize.
+
+ Parameters
+ ----------
+ filters : `list` [`int`]
+ List of input channels to each convolutional block. The length of
+ ``filters`` determines the depth of the encoder. The first element
+ of ``filters`` has to be the number of channels of the input
+ images.
+ block : :py:class:`pysegcnn.core.layers.EncoderBlock`
+ The convolutional block defining a layer in the encoder.
+ A subclass of :py:class:`pysegcnn.core.layers.EncoderBlock`, e.g.
+ :py:class:`pysegcnn.core.layers.ConvBnReluMaxPool`.
+ **kwargs: `dict` [`str`]
+ Additional arguments passed to
+ :py:class:`pysegcnn.core.layers.Conv2dSame`.
+
+ """
super().__init__()
# the number of filters for each block: the first element of filters
@@ -423,22 +422,22 @@ class Encoder(nn.Module):
"""Forward pass of the encoder.
Stores intermediate outputs in a dictionary. The keys of the dictionary
- are the number of the network layers and the values are dictionaries
+ are the numbers of the network layers and the values are dictionaries
with the following (key, value) pairs:
- ``"feature"``
- The intermediate encoder outputs (`torch.Tensor`).
- ``"indices"``
+ ``'feature'``
+ The intermediate encoder outputs (:py:class:`torch.Tensor`).
+ ``'indices'``
The indices of the max pooling layer, if required
- (`torch.Tensor`).
+ (:py:class:`torch.Tensor`).
Parameters
----------
- x : `torch.Tensor`, shape=(batch, channel, height, width)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h, w)
Input image.
Returns
-------
- x : `torch.Tensor`, shape=(batch, channel, height, width)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h, w)
Output of the encoder.
"""
@@ -459,33 +458,47 @@ class Decoder(nn.Module):
"""Generic convolutional decoder.
When instanciating an encoder-decoder architechure, ``filters`` should be
- the same for `pysegcnn.core.layers.Encoder` and
- `pysegcnn.core.layers.Decoder`.
+ the same for :py:class:`pysegcnn.core.layers.Encoder` and
+ :py:class:`pysegcnn.core.layers.Decoder`.
- See `pysegcnn.core.models.UNet` for an example implementation.
+ See :py:class:`pysegcnn.core.models.UNet` for an example implementation.
- Parameters
+ Attributes
----------
- filters : `list` [`int`]
- List of input channels to each convolutional block. The length of
- ``filters`` determines the depth of the decoder. The first element of
- ``filters`` has to be the number of channels of the input images.
- block : `pysegcnn.core.layers.DecoderBlock`
+ features : :py:class:`numpy.ndarray`
+ Input channels to each convolutional block, i.e. ``filters``.
+ block : :py:class:`pysegcnn.core.layers.EncoderBlock`
The convolutional block defining a layer in the decoder.
- A subclass of `pysegcnn.core.layers.DecoderBlock`, e.g.
- `pysegcnn.core.layers.ConvBnReluMaxUnpool`.
+ A subclass of :py:class:`pysegcnn.core.layers.EncoderBlock`, e.g.
+ :py:class:`pysegcnn.core.layers.ConvBnReluMaxPool`.
skip : `bool`
Whether to apply skip connections from the encoder to the decoder.
- **kwargs: 'dict' [`str`]
- Additional arguments passed to `pysegcnn.core.layers.Conv2dSame`.
-
- Returns
- -------
- None.
+ layers : :py:class:`torch.nn.ModuleList`
+ List of blocks in the decoder.
"""
def __init__(self, filters, block, skip=True, **kwargs):
+ """Initialize.
+
+ Parameters
+ ----------
+ filters : `list` [`int`]
+ List of input channels to each convolutional block. The length of
+ ``filters`` determines the depth of the decoder. The first element
+ of ``filters`` has to be the number of channels of the input
+ images.
+ block : :py:class:`pysegcnn.core.layers.DecoderBlock`
+ The convolutional block defining a layer in the decoder.
+ A subclass of :py:class:`pysegcnn.core.layers.DecoderBlock`, e.g.
+ :py:class:`pysegcnn.core.layers.ConvBnReluMaxUnpool`.
+ skip : `bool`, optional
+ Whether to apply skip connections from the encoder to the decoder.
+ **kwargs: `dict` [`str`]
+ Additional arguments passed to
+ :py:class:`pysegcnn.core.layers.Conv2dSame`.
+
+ """
super().__init__()
# the block of operations defining a layer in the decoder
@@ -520,20 +533,22 @@ class Decoder(nn.Module):
Parameters
----------
- x : `torch.Tensor`, shape=(batch, channel, height, width)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h, w)
Output of the encoder.
enc_cache : `dict` [`dict`]
Cache dictionary. The keys of the dictionary are the number of the
network layers and the values are dictionaries with the following
(key, value) pairs:
- ``"feature"``
- The intermediate encoder outputs (`torch.Tensor`).
- ``"indices"``
- The indices of the max pooling layer (`torch.Tensor`).
+ ``'feature'``
+ The intermediate encoder outputs
+ (:py:class:`torch.Tensor`).
+ ``'indices'``
+ The indices of the max pooling layer
+ (:py:class:`torch.Tensor`).
Returns
-------
- x : `torch.Tensor`, shape=(batch, channel, height, width)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h, w)
Output of the decoder.
"""
@@ -552,23 +567,7 @@ class Decoder(nn.Module):
class ConvBnReluMaxPool(EncoderBlock):
- """Block of convolution, batchnorm, relu and 2x2 max pool.
-
- Parameters
- ----------
- in_channels : `int`
- Number of input channels.
- out_channels : `int`
- Number of output channels.
- **kwargs: 'dict' [`str`]
- Additional keyword arguments passed to
- `pysegcnn.core.layers.Conv2dSame`.
-
- Returns
- -------
- None.
-
- """
+ """Block of convolution, batchnorm, relu and 2x2 max pool."""
def __init__(self, in_channels, out_channels, **kwargs):
super().__init__(in_channels, out_channels, **kwargs)
@@ -578,29 +577,30 @@ class ConvBnReluMaxPool(EncoderBlock):
Returns
-------
- layers : `torch.nn.Sequential` [`torch.nn.Module`]
- An instance of `torch.nn.Sequential` containing the sequence
- of convolution, batchnorm and relu layer (`torch.nn.Module`)
- instances.
+ layers : :py:class:`torch.nn.Sequential` [:py:class:`torch.nn.Module`]
+ An instance of :py:class:`torch.nn.Sequential` containing the
+ sequence of convolution, batchnorm and relu layer
+ (:py:class:`torch.nn.Module`) instances.
"""
return conv_bn_relu(self.in_channels, self.out_channels, **self.kwargs)
def downsample(self, x):
- """2x2 max pooling layer, `torch.nn.functional.max_pool2d`.
+ """2x2 max pooling layer.
Parameters
----------
- x : `torch.Tensor`, shape=(batch, channel, height, width)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h, w)
Input tensor.
Returns
-------
- x : `torch.Tensor`, shape=(batch, channel, height // 2, width // 2)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h // 2, w // 2)
The 2x2 max pooled tensor.
- indices : `torch.Tensor` or `None`
+ indices : :py:class:`torch.Tensor` or `None`
The indices of the maxima. Useful for upsampling with
- `torch.nn.functional.max_unpool2d`.
+ :py:func:`torch.nn.functional.max_unpool2d`.
+
"""
x, indices = F.max_pool2d(x, kernel_size=2, return_indices=True)
return x, indices
@@ -618,23 +618,7 @@ class ConvBnReluMaxPool(EncoderBlock):
class ConvBnReluMaxUnpool(DecoderBlock):
- """Block of convolution, batchnorm, relu and 2x2 max unpool.
-
- Parameters
- ----------
- in_channels : `int`
- Number of input channels.
- out_channels : `int`
- Number of output channels
- **kwargs: 'dict' [`str`]
- Additional keyword arguments passed to
- `pysegcnn.core.layers.Conv2dSame`.
-
- Returns
- -------
- None.
-
- """
+ """Block of convolution, batchnorm, relu and 2x2 max unpool."""
def __init__(self, in_channels, out_channels, **kwargs):
super().__init__(in_channels, out_channels, **kwargs)
@@ -644,10 +628,10 @@ class ConvBnReluMaxUnpool(DecoderBlock):
Returns
-------
- layers : `torch.nn.Sequential` [`torch.nn.Module`]
- An instance of `torch.nn.Sequential` containing the sequence
- of convolution, batchnorm and relu layer (`torch.nn.Module`)
- instances.
+ layers : :py:class:`torch.nn.Sequential` [:py:class:`torch.nn.Module`]
+ An instance of :py:class:`torch.nn.Sequential` containing the
+ sequence of convolution, batchnorm and relu layer
+ (:py:class:`torch.nn.Module`) instances.
"""
return conv_bn_relu(self.in_channels, self.out_channels, **self.kwargs)
@@ -657,18 +641,18 @@ class ConvBnReluMaxUnpool(DecoderBlock):
Parameters
----------
- x : `torch.Tensor`, shape=(batch, channel, height, width)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h, w)
Input tensor.
- feature : `torch.Tensor`, shape=(batch, channel, height, width)
+ feature : :py:class:`torch.Tensor`, shape=(b, c, h, w)
Intermediate output of a layer in the encoder. Used to determine
the output shape of the upsampling operation.
- indices : `torch.Tensor`
+ indices : :py:class:`torch.Tensor`
The indices of the maxima of the max pooling operation
- (as returned by `torch.nn.functional.max_pool2d`).
+ (as returned by :py:func:`torch.nn.functional.max_pool2d`).
Returns
-------
- x : `torch.Tensor`, shape=(batch, channel, height * 2, width * 2)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h * 2, w * 2)
The 2x2 max unpooled tensor.
"""
@@ -688,22 +672,7 @@ class ConvBnReluMaxUnpool(DecoderBlock):
class ConvBnReluUpsample(DecoderBlock):
- """Block of convolution, batchnorm, relu and nearest neighbor upsampling.
-
- Parameters
- ----------
- in_channels : `int`
- Number of input channels.
- out_channels : `int`
- Number of output channels
- **kwargs: 'dict' [`str`]
- Additional arguments passed to `pysegcnn.core.layers.Conv2dSame`.
-
- Returns
- -------
- None.
-
- """
+ """Block of convolution, batchnorm, relu and nearest neighbor upsample."""
def __init__(self, in_channels, out_channels, **kwargs):
super().__init__(in_channels, out_channels, **kwargs)
@@ -713,10 +682,10 @@ class ConvBnReluUpsample(DecoderBlock):
Returns
-------
- layers : `torch.nn.Sequential` [`torch.nn.Module`]
- An instance of `torch.nn.Sequential` containing the sequence
- of convolution, batchnorm and relu layer (`torch.nn.Module`)
- instances.
+ layers : :py:class:`torch.nn.Sequential` [:py:class:`torch.nn.Module`]
+ An instance of :py:class:`torch.nn.Sequential` containing the
+ sequence of convolution, batchnorm and relu layer
+ (:py:class:`torch.nn.Module`) instances.
"""
return conv_bn_relu(self.in_channels, self.out_channels, **self.kwargs)
@@ -726,19 +695,19 @@ class ConvBnReluUpsample(DecoderBlock):
Parameters
----------
- x : `torch.Tensor`, shape=(batch, channel, height, width)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h, w)
Input tensor.
- feature : `torch.Tensor`, shape=(batch, channel, height, width)
+ feature : :py:class:`torch.Tensor`, shape=(b, c, h, w)
Intermediate output of a layer in the encoder. Used to determine
the output shape of the upsampling operation.
indices : `None`, optional
The indices of the maxima of the max pooling operation
- (as returned by `torch.nn.functional.max_pool2d`). Not required by
- this upsampling method.
+ (as returned by :py:func:`torch.nn.functional.max_pool2d`).
+ Not required by this upsampling method.
Returns
-------
- x : `torch.Tensor`, shape=(batch, channel, height, width)
+ x : :py:class:`torch.Tensor`, shape=(b, c, h, w)
The 2x2 upsampled tensor.
"""
diff --git a/pysegcnn/core/models.py b/pysegcnn/core/models.py
index 39ba1a1c4d6247e67aa0ef7a028e992ff53584c9..e24c76c8d1ce528fc8ea2348a63471d2ae0ff07b 100644
--- a/pysegcnn/core/models.py
+++ b/pysegcnn/core/models.py
@@ -37,15 +37,17 @@ class Network(nn.Module):
"""Generic Network class.
The base class for each model. If you want to implement a new model,
- inherit the ``~pysegcnn.core.models.Network`` class.
+ inherit the :py:class:`pysegcnn.core.models.Network` class.
- Returns
- -------
- None.
+ Attributes
+ ----------
+ state_file : `str` or `None` or :py:class:`pathlib.Path`
+ The model state file, where the model parameters are saved.
"""
def __init__(self):
+ """Initialize."""
super().__init__()
# initialize state file
@@ -89,9 +91,9 @@ class Network(nn.Module):
Parameters
----------
- state_file : `str` or `pathlib.Path`
+ state_file : `str` or :py:class:`pathlib.Path`
Path to save the model state.
- optimizer : `torch.optim.Optimizer`
+ optimizer : :py:class:`torch.optim.Optimizer`
The optimizer used to train the model.
bands : `list` [`str`] or `None`, optional
List of bands the model is trained with. The default is None.
@@ -102,7 +104,7 @@ class Network(nn.Module):
Returns
-------
model_state : `dict`
- A dictionary containing the model and optimizer state
+ A dictionary containing the model and optimizer state.
"""
# check if the output path exists and if not, create it
@@ -156,7 +158,7 @@ class Network(nn.Module):
Parameters
----------
- state_file : `str` or `pathlib.Path`
+ state_file : `str` or :py:class:`pathlib.Path`
The model state file. Model state files are stored in
pysegcnn/main/_models.
@@ -167,13 +169,13 @@ class Network(nn.Module):
Returns
-------
- model : `pysegcnn.core.models.Network`
+ model : :py:class:`pysegcnn.core.models.Network`
The pretrained model.
- optimizer : `torch.optim.Optimizer`
+ optimizer : :py:class:`torch.optim.Optimizer`
The optimizer used to train the model.
- model_state : '`dict`
+ model_state : `dict`
A dictionary containing the model and optimizer state, as
- constructed by `~pysegcnn.core.Network.save`.
+ constructed by :py:meth:`~pysegcnn.core.Network.save`.
"""
# load the pretrained model
@@ -216,7 +218,7 @@ class Network(nn.Module):
Returns
-------
- state_file : `pathlib.Path` or `None`
+ state_file : :py:class:`pathlib.Path` or `None`
The model state file.
"""
@@ -224,36 +226,58 @@ class Network(nn.Module):
class UNet(Network):
- """A PyTorch implementation of `U-Net`_.
+ """A slightly modified implementation of `U-Net`_ in PyTorch.
+
+ .. important::
- Slightly modified version of U-Net:
- - each convolution is followed by a batch normalization layer
- - the upsampling is implemented by a 2x2 max unpooling operation
+ - Each convolution is followed by a batch normalization layer
+ - Upsampling is implemented by a 2x2 max unpooling operation
.. _U-Net:
https://arxiv.org/abs/1505.04597
- Parameters
+ Attributes
----------
in_channels : `int`
Number of channels of the input images.
nclasses : `int`
Number of classes.
- filters : `list` [`int`]
+ kwargs : `dict` [`str`]
+ Additional keyword arguments passed to
+ :py:class:`pysegcnn.core.layers.Conv2dSame`.
+ nfilters : `list` [`int`]
List of input channels to each convolutional block.
skip : `bool`
Whether to apply skip connections from the encoder to the decoder.
- **kwargs: 'dict' [`str`]
- Additional keyword arguments passed to
- `pysegcnn.core.layers.Conv2dSame`.
-
- Returns
- -------
- None.
+ epoch : `int`
+ Number of epochs the model was trained.
+ encoder : :py:class:`pysegcnn.core.layers.Encoder`
+ The convolutional encoder.
+ decoder : :py:class:`pysegcnn.core.layers.Decoder`
+ The convolutional decoder.
+ classifier : :py:class:`pysegcnn.core.layers.Conv2dSame`
+ The classification layer, a 1x1 convolution.
"""
def __init__(self, in_channels, nclasses, filters, skip, **kwargs):
+ """Initialize.
+
+ Parameters
+ ----------
+ in_channels : `int`
+ Number of channels of the input images.
+ nclasses : `int`
+ Number of classes.
+ filters : `list` [`int`]
+ List of input channels to each convolutional block.
+ skip : `bool`
+ Whether to apply skip connections from the encoder to the decoder.
+ **kwargs: `dict` [`str`]
+ Additional keyword arguments passed to
+ :py:class:`pysegcnn.core.layers.Conv2dSame`.
+
+ """
super().__init__()
# number of input channels