################################################################################
# Copyright (c) 2021 ContinualAI. #
# Copyrights licensed under the MIT License. #
# See the accompanying LICENSE file for terms. #
# #
# Date: 20-05-2020 #
# Author: Matthias De Lange #
# E-mail: contact@continualai.org #
# Website: continualai.org #
################################################################################
from pathlib import Path
from typing import Union, Any, Optional
from avalanche.benchmarks.classic.classic_benchmarks_utils import (
check_vision_benchmark,
)
from avalanche.benchmarks.datasets import (
INATURALIST2018,
default_dataset_location,
)
from avalanche.benchmarks import nc_benchmark
from torchvision import transforms
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
_default_train_transform = transforms.Compose(
[
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize,
]
)
_default_eval_transform = transforms.Compose(
[
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize,
]
)
[docs]def SplitInaturalist(
*,
super_categories=None,
return_task_id=False,
download=False,
seed=0,
class_ids_from_zero_in_each_exp: bool = False,
class_ids_from_zero_from_first_exp: bool = False,
train_transform: Optional[Any] = _default_train_transform,
eval_transform: Optional[Any] = _default_eval_transform,
dataset_root: Optional[Union[str, Path]] = None
):
"""Creates a CL benchmark using the iNaturalist2018 dataset.
A selection of supercategories (by default 10) define the experiences.
Note that the supercategories are highly imbalanced in the number of classes
and the amount of data available.
If the dataset is not present in the computer, **this method will
automatically download** and store it if `download=True`
(120Gtrain/val).
To parse the dataset jsons you need to install an additional dependency:
"pycocotools". You can install it with the command
``conda install -c conda-forge pycocotools``
Implementation is based on the CL survey
(https://ieeexplore.ieee.org/document/9349197) but differs slightly.
The survey uses only the original iNaturalist2018 training dataset split
into 70/10/20 for train/val/test streams. This method instead uses the full
iNaturalist2018 training set to make the `train_stream`, whereas the
`test_stream` is defined by the original iNaturalist2018 validation data.
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.
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 super_categories: The list of supercategories which define the
tasks, i.e. each task consists of all classes in a super-category.
:param download: If true and the dataset is not present in the computer,
this method will automatically download and store it. This will take
120G for the train/val set.
: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 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
'inatuarlist2018' will be used.
:returns: A properly initialized :class:`NCScenario` instance.
"""
# Categories with > 100 datapoints
if super_categories is None:
super_categories = [
"Amphibia",
"Animalia",
"Arachnida",
"Aves",
"Fungi",
"Insecta",
"Mammalia",
"Mollusca",
"Plantae",
"Reptilia",
]
train_set, test_set = _get_inaturalist_dataset(
dataset_root, super_categories, download=download
)
per_exp_classes, fixed_class_order = _get_split(super_categories, train_set)
return nc_benchmark(
fixed_class_order=fixed_class_order,
per_exp_classes=per_exp_classes,
train_dataset=train_set,
test_dataset=test_set,
n_experiences=len(super_categories),
task_labels=return_task_id,
seed=seed,
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_inaturalist_dataset(dataset_root, super_categories, download):
if dataset_root is None:
dataset_root = default_dataset_location("inatuarlist2018")
train_set = INATURALIST2018(
str(dataset_root), split="train", supcats=super_categories, download=download
)
test_set = INATURALIST2018(
str(dataset_root), split="val", supcats=super_categories, download=download
)
return train_set, test_set
def _get_split(super_categories, train_set):
"""Get number of classes per experience, and
the total order of the classes."""
per_exp_classes, fixed_class_order = {}, []
for idx, supcat in enumerate(super_categories):
new_cats = list(train_set.cats_per_supcat[supcat])
fixed_class_order += new_cats
per_exp_classes[idx] = len(new_cats)
return per_exp_classes, fixed_class_order
__all__ = ["SplitInaturalist"]
if __name__ == "__main__":
import sys
benchmark_instance = SplitInaturalist()
check_vision_benchmark(benchmark_instance, show_without_transforms=False)
sys.exit(0)