5399886876

# The dataset code has been adapted from:
# https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html
# from https://github.com/pytorch/tutorials
# which has been distributed under the following license:
################################################################################
# BSD 3-Clause License
#
# Copyright (c) 2017, Pytorch contributors
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are met:
#
# * Redistributions of source code must retain the above copyright notice, this
#   list of conditions and the following disclaimer.
#
# * Redistributions in binary form must reproduce the above copyright notice,
#   this list of conditions and the following disclaimer in the documentation
#   and/or other materials provided with the distribution.
#
# * Neither the name of the copyright holder nor the names of its
#   contributors may be used to endorse or promote products derived from
#   this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
################################################################################

# For the Avalanche data loader adaptation:
################################################################################
# Copyright (c) 2022 ContinualAI                                               #
# Copyrights licensed under the MIT License.                                   #
# See the accompanying LICENSE file for terms.                                 #
#                                                                              #
# Date: 21-03-2022                                                             #
# Author: Lorenzo Pellegrini                                                   #
#                                                                              #
# E-mail: contact@continualai.org                                              #
# Website: www.continualai.org                                                 #
################################################################################


from pathlib import Path
from typing import Dict, List, Optional, Sequence, Union

import numpy as np
import torch
from PIL import Image
from torchvision.datasets.folder import default_loader

from avalanche.benchmarks.datasets import (
    SimpleDownloadableDataset,
    default_dataset_location,
)
from avalanche.benchmarks.datasets.penn_fudan.penn_fudan_data import (
    penn_fudan_data,
)


def default_mask_loader(mask_path):
    return Image.open(mask_path)


class PennFudanDataset(SimpleDownloadableDataset):
    """
    The Penn-Fudan Pedestrian detection and segmentation dataset

    Adapted from the "TorchVision Object Detection Finetuning Tutorial":
    https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html
    """

    def __init__(
        self,
        root: Optional[Union[str, Path]] = None,
        *,
        transform=None,
        loader=default_loader,
        mask_loader=default_mask_loader,
        download=True
    ):
        """
        Creates an instance of the Penn-Fudan dataset.

        :param root: The directory where the dataset can be found or downloaded.
            Defaults to None, which means that the default location for
            "pennfudanped" will be used.
        :param transform: The transformation to apply to (img, annotations)
            values.
        :param loader: The image loader to use.
        :param mask_loader: The mask image loader to use.
        :param download: If True, the dataset will be downloaded if needed.
        """

        if root is None:
            root = default_dataset_location("pennfudanped")

        self.imgs: Sequence[Path] = None  # type: ignore
        self.masks: Sequence[Path] = None  # type: ignore
        self.targets: List[Dict] = None  # type: ignore
        self.transform = transform
        self.loader = loader
        self.mask_loader = mask_loader

        super().__init__(
            root,
            penn_fudan_data[0],
            penn_fudan_data[1],
            download=download,
            verbose=True,
        )

        self._load_dataset()

    def _load_metadata(self):
        # load all image files, sorting them to
        # ensure that they are aligned
        imgs = (self.root / "PennFudanPed" / "PNGImages").iterdir()
        masks = (self.root / "PennFudanPed" / "PedMasks").iterdir()

        self.imgs = list(sorted(imgs))
        self.masks = list(sorted(masks))

        self.targets = [self.make_targets(i) for i in range(len(self.imgs))]
        return Path(self.imgs[0]).exists() and Path(self.masks[0]).exists()

    def make_targets(self, idx):
        # load images and masks
        mask_path = self.masks[idx]

        # note that we haven't converted the mask to RGB,
        # because each color corresponds to a different instance
        # with 0 being background
        mask = self.mask_loader(mask_path)
        # convert the PIL Image into a numpy array
        mask = np.array(mask)
        # instances are encoded as different colors
        obj_ids = np.unique(mask)
        # first id is the background, so remove it
        obj_ids = obj_ids[1:]

        # split the color-encoded mask into a set
        # of binary masks
        masks = mask == obj_ids[:, None, None]

        # get bounding box coordinates for each mask
        num_objs = len(obj_ids)
        boxes = []
        for i in range(num_objs):
            pos = np.where(masks[i])
            xmin: np.integer = np.min(pos[1])
            xmax: np.integer = np.max(pos[1])
            ymin: np.integer = np.min(pos[0])
            ymax: np.integer = np.max(pos[0])
            boxes.append([xmin, ymin, xmax, ymax])

        # convert everything into a torch.Tensor
        boxes_as_tensor = torch.as_tensor(boxes, dtype=torch.float32)
        del boxes
        # there is only one class
        labels = torch.ones((num_objs,), dtype=torch.int64)
        masks = torch.as_tensor(masks, dtype=torch.uint8)

        image_id = torch.tensor([idx])
        area = (boxes_as_tensor[:, 3] - boxes_as_tensor[:, 1]) * (
            boxes_as_tensor[:, 2] - boxes_as_tensor[:, 0]
        )
        # suppose all instances are not crowd
        iscrowd = torch.zeros((num_objs,), dtype=torch.int64)

        target = {}
        target["boxes"] = boxes_as_tensor
        target["labels"] = labels
        target["masks"] = masks
        target["image_id"] = image_id
        target["area"] = area
        target["iscrowd"] = iscrowd
        return target

    def __getitem__(self, idx):
        target = self.targets[idx]
        img_path = self.imgs[idx]
        img = self.loader(img_path)

        if self.transform is not None:
            img, target = self.transform(img, target)

        return img, target

    def __len__(self):
        return len(self.imgs)


if __name__ == "__main__":
    # this little example script can be used to visualize the first image
    # loaded from the dataset.
    from torch.utils.data.dataloader import DataLoader
    import matplotlib.pyplot as plt
    from torchvision import transforms
    import torch

    train_data = PennFudanDataset(
        transform=lambda im, ann: (transforms.ToTensor()(im), ann)
    )
    dataloader = DataLoader(train_data, batch_size=1)

    for batch_data in dataloader:
        x, y = batch_data
        plt.imshow(transforms.ToPILImage()(torch.squeeze(x)))
        plt.show()
        print(x.shape)
        print(y)
        break


__all__ = ["PennFudanDataset"]

1	# The dataset code has been adapted from:
2	# https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html
3	# from https://github.com/pytorch/tutorials
4	# which has been distributed under the following license:
5	################################################################################
6	# BSD 3-Clause License
7	#
8	# Copyright (c) 2017, Pytorch contributors
9	# All rights reserved.
10	#
11	# Redistribution and use in source and binary forms, with or without
12	# modification, are permitted provided that the following conditions are met:
13	#
14	# * Redistributions of source code must retain the above copyright notice, this
15	# list of conditions and the following disclaimer.
16	#
17	# * Redistributions in binary form must reproduce the above copyright notice,
18	# this list of conditions and the following disclaimer in the documentation
19	# and/or other materials provided with the distribution.
20	#
21	# * Neither the name of the copyright holder nor the names of its
22	# contributors may be used to endorse or promote products derived from
23	# this software without specific prior written permission.
24	#
25	# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
26	# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27	# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
28	# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
29	# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30	# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
31	# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
32	# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
33	# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
34	# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35	################################################################################
36
37	# For the Avalanche data loader adaptation:
38	################################################################################
39	# Copyright (c) 2022 ContinualAI #
40	# Copyrights licensed under the MIT License. #
41	# See the accompanying LICENSE file for terms. #
42	# #
43	# Date: 21-03-2022 #
44	# Author: Lorenzo Pellegrini #
45	# #
46	# E-mail: contact@continualai.org #
47	# Website: www.continualai.org #
48	################################################################################
49
50
51	from pathlib import Path	4✔
52	from typing import Dict, List, Optional, Sequence, Union	4✔
53
54	import numpy as np	4✔
55	import torch	4✔
56	from PIL import Image	4✔
57	from torchvision.datasets.folder import default_loader	4✔
58
59	from avalanche.benchmarks.datasets import (	4✔
60	SimpleDownloadableDataset,
61	default_dataset_location,
62	)
63	from avalanche.benchmarks.datasets.penn_fudan.penn_fudan_data import (	4✔
64	penn_fudan_data,
65	)
66
67
68	def default_mask_loader(mask_path):	4✔
69	return Image.open(mask_path)	×
70
71
72	class PennFudanDataset(SimpleDownloadableDataset):	4✔
73	"""	4✔
74	The Penn-Fudan Pedestrian detection and segmentation dataset
75
76	Adapted from the "TorchVision Object Detection Finetuning Tutorial":
77	https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html
78	"""
79
80	def __init__(	4✔
81	self,
82	root: Optional[Union[str, Path]] = None,
83	*,
84	transform=None,
85	loader=default_loader,
86	mask_loader=default_mask_loader,
87	download=True
88	):
89	"""
90	Creates an instance of the Penn-Fudan dataset.
91
92	:param root: The directory where the dataset can be found or downloaded.
93	Defaults to None, which means that the default location for
94	"pennfudanped" will be used.
95	:param transform: The transformation to apply to (img, annotations)
96	values.
97	:param loader: The image loader to use.
98	:param mask_loader: The mask image loader to use.
99	:param download: If True, the dataset will be downloaded if needed.
100	"""
101
102	if root is None:	×
103	root = default_dataset_location("pennfudanped")	×
104
105	self.imgs: Sequence[Path] = None # type: ignore	×
106	self.masks: Sequence[Path] = None # type: ignore	×
107	self.targets: List[Dict] = None # type: ignore	×
108	self.transform = transform	×
109	self.loader = loader	×
110	self.mask_loader = mask_loader	×
111
112	super().__init__(	×
113	root,
114	penn_fudan_data[0],
115	penn_fudan_data[1],
116	download=download,
117	verbose=True,
118	)
119
120	self._load_dataset()	×
121
122	def _load_metadata(self):	4✔
123	# load all image files, sorting them to
124	# ensure that they are aligned
125	imgs = (self.root / "PennFudanPed" / "PNGImages").iterdir()	×
126	masks = (self.root / "PennFudanPed" / "PedMasks").iterdir()	×
127
128	self.imgs = list(sorted(imgs))	×
129	self.masks = list(sorted(masks))	×
130
131	self.targets = [self.make_targets(i) for i in range(len(self.imgs))]	×
132	return Path(self.imgs[0]).exists() and Path(self.masks[0]).exists()	×
133
134	def make_targets(self, idx):	4✔
135	# load images and masks
136	mask_path = self.masks[idx]	×
137
138	# note that we haven't converted the mask to RGB,
139	# because each color corresponds to a different instance
140	# with 0 being background
141	mask = self.mask_loader(mask_path)	×
142	# convert the PIL Image into a numpy array
143	mask = np.array(mask)	×
144	# instances are encoded as different colors
145	obj_ids = np.unique(mask)	×
146	# first id is the background, so remove it
147	obj_ids = obj_ids[1:]	×
148
149	# split the color-encoded mask into a set
150	# of binary masks
151	masks = mask == obj_ids[:, None, None]	×
152
153	# get bounding box coordinates for each mask
154	num_objs = len(obj_ids)	×
155	boxes = []	×
156	for i in range(num_objs):	×
157	pos = np.where(masks[i])	×
158	xmin: np.integer = np.min(pos[1])	×
159	xmax: np.integer = np.max(pos[1])	×
160	ymin: np.integer = np.min(pos[0])	×
161	ymax: np.integer = np.max(pos[0])	×
162	boxes.append([xmin, ymin, xmax, ymax])	×
163
164	# convert everything into a torch.Tensor
165	boxes_as_tensor = torch.as_tensor(boxes, dtype=torch.float32)	×
166	del boxes	×
167	# there is only one class
168	labels = torch.ones((num_objs,), dtype=torch.int64)	×
169	masks = torch.as_tensor(masks, dtype=torch.uint8)	×
170
171	image_id = torch.tensor([idx])	×
NEW 172	area = (boxes_as_tensor[:, 3] - boxes_as_tensor[:, 1]) * (	×
173	boxes_as_tensor[:, 2] - boxes_as_tensor[:, 0]
174	)
175	# suppose all instances are not crowd
176	iscrowd = torch.zeros((num_objs,), dtype=torch.int64)	×
177
178	target = {}	×
179	target["boxes"] = boxes_as_tensor	×
180	target["labels"] = labels	×
181	target["masks"] = masks	×
182	target["image_id"] = image_id	×
183	target["area"] = area	×
184	target["iscrowd"] = iscrowd	×
185	return target	×
186
187	def __getitem__(self, idx):	4✔
188	target = self.targets[idx]	×
189	img_path = self.imgs[idx]	×
190	img = self.loader(img_path)	×
191
192	if self.transform is not None:	×
193	img, target = self.transform(img, target)	×
194
195	return img, target	×
196
197	def __len__(self):	4✔
198	return len(self.imgs)	×
199
200
201	if __name__ == "__main__":	4✔
202	# this little example script can be used to visualize the first image
203	# loaded from the dataset.
204	from torch.utils.data.dataloader import DataLoader	×
205	import matplotlib.pyplot as plt	×
206	from torchvision import transforms	×
207	import torch	×
208
209	train_data = PennFudanDataset(	×
210	transform=lambda im, ann: (transforms.ToTensor()(im), ann)
211	)
212	dataloader = DataLoader(train_data, batch_size=1)	×
213
214	for batch_data in dataloader:	×
215	x, y = batch_data	×
216	plt.imshow(transforms.ToPILImage()(torch.squeeze(x)))	×
217	plt.show()	×
218	print(x.shape)	×
219	print(y)	×
220	break	×
221
222
223	__all__ = ["PennFudanDataset"]	4✔

ContinualAI / avalanche / 5399886876

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous