################################################################################
# Copyright (c) 2021 ContinualAI. #
# Copyrights licensed under the MIT License. #
# See the accompanying LICENSE file for terms. #
# #
# Date: 13-02-2021 #
# Author(s): Jary Pomponi, Antonio Carta #
################################################################################
from pathlib import Path
from typing import Optional, Sequence, Any, Union
from torch import Tensor
from torchvision.transforms import (
ToTensor,
Compose,
Normalize,
ToPILImage,
)
from PIL.Image import Image
from avalanche.benchmarks import nc_benchmark
from avalanche.benchmarks.classic.classic_benchmarks_utils import (
check_vision_benchmark,
)
from avalanche.benchmarks.datasets import default_dataset_location
from avalanche.benchmarks.datasets.omniglot import Omniglot
_default_omniglot_train_transform = Compose(
[ToTensor(), Normalize((0.9221,), (0.2681,))]
)
_default_omniglot_eval_transform = Compose(
[ToTensor(), Normalize((0.9221,), (0.2681,))]
)
class PixelsPermutation(object):
"""Apply a fixed permutation to the pixels of the given image.
Works with both Tensors and PIL images. Returns an object of the same type
of the input element.
"""
def __init__(self, index_permutation: Sequence[int]):
self.permutation = index_permutation
self._to_tensor = ToTensor()
self._to_image = ToPILImage()
def __call__(self, img: Union[Image, Tensor]):
is_image = isinstance(img, Image)
if (not is_image) and (not isinstance(img, Tensor)):
raise ValueError("Invalid input: must be a PIL image or a Tensor")
image_as_tensor: Tensor
if is_image:
image_as_tensor = self._to_tensor(img)
else:
image_as_tensor = img
image_as_tensor = image_as_tensor.view(-1)[self.permutation].view(
*image_as_tensor.shape
)
if is_image:
img = self._to_image(image_as_tensor)
else:
img = image_as_tensor
return img
def SplitAlphabetOmniglot(
n_experiences: int,
*,
return_task_id=False,
seed: Optional[int] = None,
fixed_class_order: Optional[Sequence[int]] = None,
class_ids_from_zero_from_first_exp: bool = False,
shuffle: bool = True,
train_transform: Optional[Any] = _default_omniglot_train_transform,
eval_transform: Optional[Any] = _default_omniglot_eval_transform,
dataset_root: Optional[Union[str, Path]] = None
):
"""Class-incremental OMNIGLOT with the alphabet used as target.
If the dataset is not present in the computer, this method will
automatically download and store it.
The returned benchmark will return experiences containing all patterns of a
subset of alphabets (class-incremental scenario).
The benchmark API is quite simple and is uniform across all benchmark
generators. It is recommended to check the tutorial of the "benchmark" API,
which contains usage examples ranging from "basic" to "advanced".
:param n_experiences: The number of incremental experiences in the current
benchmark. The value of this parameter should be a divisor of 10.
:param return_task_id: if True, a progressive task id is returned for every
experience. If False, all experiences will have a task ID of 0.
:param seed: A valid int used to initialize the random number generator.
Can be None.
:param fixed_class_order: A list of class IDs used to define the class
order. If None, value of ``seed`` will be used to define the class
order. If non-None, ``seed`` parameter will be ignored.
Defaults to None.
:param class_ids_from_zero_from_first_exp: If True, original class IDs
will be remapped so that they will appear as having an ascending
order. For instance, if the resulting class order after shuffling
(or defined by fixed_class_order) is [23, 34, 11, 7, 6, ...] and
class_ids_from_zero_from_first_exp is True, then all the patterns
belonging to class 23 will appear as belonging to class "0",
class "34" will be mapped to "1", class "11" to "2" and so on.
This is very useful when drawing confusion matrices and when dealing
with algorithms with dynamic head expansion. Defaults to False.
Mutually exclusive with the ``class_ids_from_zero_in_each_exp``
parameter.
:param shuffle: If true, the class order in the incremental experiences is
randomly shuffled. Default to True.
:param train_transform: The transformation to apply to the training data,
e.g. a random crop, a normalization or a concatenation of different
transformations (see torchvision.transform documentation for a
comprehensive list of possible transformations).
If no transformation is passed, the default train transformation
will be used.
:param eval_transform: The transformation to apply to the test data,
e.g. a random crop, a normalization or a concatenation of different
transformations (see torchvision.transform documentation for a
comprehensive list of possible transformations).
If no transformation is passed, the default test transformation
will be used.
:param dataset_root: The root path of the dataset. Defaults to None, which
means that the default location for 'omniglot' will be used.
:returns: A properly initialized :class:`NCScenario` instance.
"""
omniglot_train, omniglot_test = _get_omniglot_dataset(dataset_root)
return nc_benchmark(
train_dataset=omniglot_train,
test_dataset=omniglot_test,
n_experiences=n_experiences,
task_labels=return_task_id,
seed=seed,
fixed_class_order=fixed_class_order,
shuffle=shuffle,
class_ids_from_zero_in_each_exp=False,
class_ids_from_zero_from_first_exp=class_ids_from_zero_from_first_exp,
train_transform=train_transform,
eval_transform=eval_transform,
)
[docs]def SplitOmniglot(
n_experiences: int,
*,
return_task_id=False,
seed: Optional[int] = None,
fixed_class_order: Optional[Sequence[int]] = None,
shuffle: bool = True,
class_ids_from_zero_in_each_exp: bool = False,
class_ids_from_zero_from_first_exp: bool = False,
train_transform: Optional[Any] = _default_omniglot_train_transform,
eval_transform: Optional[Any] = _default_omniglot_eval_transform,
dataset_root: Optional[Union[str, Path]] = None
):
"""
Creates a CL benchmark using the OMNIGLOT dataset.
If the dataset is not present in the computer, this method will
automatically download and store it.
The returned benchmark will return experiences containing all patterns of a
subset of classes, which means that each class is only seen "once".
This is one of the most common scenarios in the Continual Learning
literature. Common names used in literature to describe this kind of
scenario are "Class Incremental", "New Classes", etc.
By default, an equal amount of classes will be assigned to each experience.
OMNIGLOT consists of 964 classes, which means that the number of
experiences can be 1, 2, 4, 241, 482, 964.
This generator doesn't force a choice on the availability of task labels,
a choice that is left to the user (see the `return_task_id` parameter for
more info on task labels).
The benchmark instance returned by this method will have two fields,
`train_stream` and `test_stream`, which can be iterated to obtain
training and test :class:`Experience`. Each Experience contains the
`dataset` and the associated task label.
The benchmark API is quite simple and is uniform across all benchmark
generators. It is recommended to check the tutorial of the "benchmark" API,
which contains usage examples ranging from "basic" to "advanced".
:param n_experiences: The number of incremental experiences in the current
benchmark. The value of this parameter should be a divisor of 10.
:param return_task_id: if True, a progressive task id is returned for every
experience. If False, all experiences will have a task ID of 0.
:param seed: A valid int used to initialize the random number generator.
Can be None.
:param fixed_class_order: A list of class IDs used to define the class
order. If None, value of ``seed`` will be used to define the class
order. If non-None, ``seed`` parameter will be ignored.
Defaults to None.
:param shuffle: If true, the class order in the incremental experiences is
randomly shuffled. Default to True.
:param class_ids_from_zero_in_each_exp: If True, original class IDs
will be mapped to range [0, n_classes_in_exp) for each experience.
Defaults to False. Mutually exclusive with the
``class_ids_from_zero_from_first_exp`` parameter.
:param class_ids_from_zero_from_first_exp: If True, original class IDs
will be remapped so that they will appear as having an ascending
order. For instance, if the resulting class order after shuffling
(or defined by fixed_class_order) is [23, 34, 11, 7, 6, ...] and
class_ids_from_zero_from_first_exp is True, then all the patterns
belonging to class 23 will appear as belonging to class "0",
class "34" will be mapped to "1", class "11" to "2" and so on.
This is very useful when drawing confusion matrices and when dealing
with algorithms with dynamic head expansion. Defaults to False.
Mutually exclusive with the ``class_ids_from_zero_in_each_exp``
parameter.
:param train_transform: The transformation to apply to the training data,
e.g. a random crop, a normalization or a concatenation of different
transformations (see torchvision.transform documentation for a
comprehensive list of possible transformations).
If no transformation is passed, the default train transformation
will be used.
:param eval_transform: The transformation to apply to the test data,
e.g. a random crop, a normalization or a concatenation of different
transformations (see torchvision.transform documentation for a
comprehensive list of possible transformations).
If no transformation is passed, the default test transformation
will be used.
:param dataset_root: The root path of the dataset. Defaults to None, which
means that the default location for 'omniglot' will be used.
:returns: A properly initialized :class:`NCScenario` instance.
"""
omniglot_train, omniglot_test = _get_omniglot_dataset(dataset_root)
return nc_benchmark(
train_dataset=omniglot_train,
test_dataset=omniglot_test,
n_experiences=n_experiences,
task_labels=return_task_id,
seed=seed,
fixed_class_order=fixed_class_order,
shuffle=shuffle,
class_ids_from_zero_in_each_exp=class_ids_from_zero_in_each_exp,
class_ids_from_zero_from_first_exp=class_ids_from_zero_from_first_exp,
train_transform=train_transform,
eval_transform=eval_transform,
)
def _get_omniglot_dataset(dataset_root):
if dataset_root is None:
dataset_root = default_dataset_location("omniglot")
train = Omniglot(root=dataset_root, train=True, download=True)
test = Omniglot(root=dataset_root, train=False, download=True)
return train, test
__all__ = ["SplitOmniglot"]
if __name__ == "__main__":
import sys
print("Split Omniglot")
benchmark_instance = SplitOmniglot(4, train_transform=None, eval_transform=None)
check_vision_benchmark(benchmark_instance)
sys.exit(0)