rawalkhirodkar's picture
Add initial commit
28c256d
raw
history blame
24.9 kB
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
from typing import Any, Callable, Dict, List, Optional, Sequence, Union
import mmengine
import numpy as np
from .base import BaseTransform
from .builder import TRANSFORMS
from .utils import cache_random_params, cache_randomness
# Define type of transform or transform config
Transform = Union[Dict, Callable[[Dict], Dict]]
# Indicator of keys marked by KeyMapper._map_input, which means ignoring the
# marked keys in KeyMapper._apply_transform so they will be invisible to
# wrapped transforms.
# This can be 2 possible case:
# 1. The key is required but missing in results
# 2. The key is manually set as ... (Ellipsis) in ``mapping``, which means
# the original value in results should be ignored
IgnoreKey = object()
# Import nullcontext if python>=3.7, otherwise use a simple alternative
# implementation.
try:
from contextlib import nullcontext # type: ignore
except ImportError:
from contextlib import contextmanager
@contextmanager # type: ignore
def nullcontext(resource=None):
try:
yield resource
finally:
pass
@TRANSFORMS.register_module()
class Compose(BaseTransform):
"""Compose multiple transforms sequentially.
Args:
transforms (list[dict | callable]): Sequence of transform object or
config dict to be composed.
Examples:
>>> pipeline = [
>>> dict(type='Compose',
>>> transforms=[
>>> dict(type='LoadImageFromFile'),
>>> dict(type='Normalize')
>>> ]
>>> )
>>> ]
"""
def __init__(self, transforms: Union[Transform, Sequence[Transform]]):
super().__init__()
if not isinstance(transforms, Sequence):
transforms = [transforms]
self.transforms: List = []
for transform in transforms:
if isinstance(transform, dict):
transform = TRANSFORMS.build(transform)
self.transforms.append(transform)
elif callable(transform):
self.transforms.append(transform)
else:
raise TypeError('transform must be callable or a dict, but got'
f' {type(transform)}')
def __iter__(self):
"""Allow easy iteration over the transform sequence."""
return iter(self.transforms)
def transform(self, results: Dict) -> Optional[Dict]:
"""Call function to apply transforms sequentially.
Args:
results (dict): A result dict contains the results to transform.
Returns:
dict or None: Transformed results.
"""
for t in self.transforms:
results = t(results) # type: ignore
if results is None:
return None
return results
def __repr__(self):
"""Compute the string representation."""
format_string = self.__class__.__name__ + '('
for t in self.transforms:
format_string += f'\n {t}'
format_string += '\n)'
return format_string
@TRANSFORMS.register_module()
class KeyMapper(BaseTransform):
"""A transform wrapper to map and reorganize the input/output of the
wrapped transforms (or sub-pipeline).
Args:
transforms (list[dict | callable], optional): Sequence of transform
object or config dict to be wrapped.
mapping (dict): A dict that defines the input key mapping.
The keys corresponds to the inner key (i.e., kwargs of the
``transform`` method), and should be string type. The values
corresponds to the outer keys (i.e., the keys of the
data/results), and should have a type of string, list or dict.
None means not applying input mapping. Default: None.
remapping (dict): A dict that defines the output key mapping.
The keys and values have the same meanings and rules as in the
``mapping``. Default: None.
auto_remap (bool, optional): If True, an inverse of the mapping will
be used as the remapping. If auto_remap is not given, it will be
automatically set True if 'remapping' is not given, and vice
versa. Default: None.
allow_nonexist_keys (bool): If False, the outer keys in the mapping
must exist in the input data, or an exception will be raised.
Default: False.
Examples:
>>> # Example 1: KeyMapper 'gt_img' to 'img'
>>> pipeline = [
>>> # Use KeyMapper to convert outer (original) field name
>>> # 'gt_img' to inner (used by inner transforms) filed name
>>> # 'img'
>>> dict(type='KeyMapper',
>>> mapping={'img': 'gt_img'},
>>> # auto_remap=True means output key mapping is the revert of
>>> # the input key mapping, e.g. inner 'img' will be mapped
>>> # back to outer 'gt_img'
>>> auto_remap=True,
>>> transforms=[
>>> # In all transforms' implementation just use 'img'
>>> # as a standard field name
>>> dict(type='Crop', crop_size=(384, 384)),
>>> dict(type='Normalize'),
>>> ])
>>> ]
>>> # Example 2: Collect and structure multiple items
>>> pipeline = [
>>> # The inner field 'imgs' will be a dict with keys 'img_src'
>>> # and 'img_tar', whose values are outer fields 'img1' and
>>> # 'img2' respectively.
>>> dict(type='KeyMapper',
>>> dict(
>>> type='KeyMapper',
>>> mapping=dict(
>>> imgs=dict(
>>> img_src='img1',
>>> img_tar='img2')),
>>> transforms=...)
>>> ]
>>> # Example 3: Manually set ignored keys by "..."
>>> pipeline = [
>>> ...
>>> dict(type='KeyMapper',
>>> mapping={
>>> # map outer key "gt_img" to inner key "img"
>>> 'img': 'gt_img',
>>> # ignore outer key "mask"
>>> 'mask': ...,
>>> },
>>> transforms=[
>>> dict(type='RandomFlip'),
>>> ])
>>> ...
>>> ]
"""
def __init__(self,
transforms: Union[Transform, List[Transform]] = None,
mapping: Optional[Dict] = None,
remapping: Optional[Dict] = None,
auto_remap: Optional[bool] = None,
allow_nonexist_keys: bool = False):
super().__init__()
self.allow_nonexist_keys = allow_nonexist_keys
self.mapping = mapping
if auto_remap is None:
auto_remap = remapping is None
self.auto_remap = auto_remap
if self.auto_remap:
if remapping is not None:
raise ValueError('KeyMapper: ``remapping`` must be None if'
'`auto_remap` is set True.')
self.remapping = mapping
else:
self.remapping = remapping
if transforms is None:
transforms = []
self.transforms = Compose(transforms)
def __iter__(self):
"""Allow easy iteration over the transform sequence."""
return iter(self.transforms)
def _map_input(self, data: Dict,
mapping: Optional[Dict]) -> Dict[str, Any]:
"""KeyMapper inputs for the wrapped transforms by gathering and
renaming data items according to the mapping.
Args:
data (dict): The original input data
mapping (dict, optional): The input key mapping. See the document
of ``mmcv.transforms.wrappers.KeyMapper`` for details. In
set None, return the input data directly.
Returns:
dict: The input data with remapped keys. This will be the actual
input of the wrapped pipeline.
"""
if mapping is None:
return data.copy()
def _map(data, m):
if isinstance(m, dict):
# m is a dict {inner_key:outer_key, ...}
return {k_in: _map(data, k_out) for k_in, k_out in m.items()}
if isinstance(m, (tuple, list)):
# m is a list or tuple [outer_key1, outer_key2, ...]
# This is the case when we collect items from the original
# data to form a list or tuple to feed to the wrapped
# transforms.
return m.__class__(_map(data, e) for e in m)
# allow manually mark a key to be ignored by ...
if m is ...:
return IgnoreKey
# m is an outer_key
if self.allow_nonexist_keys:
return data.get(m, IgnoreKey)
else:
return data.get(m)
collected = _map(data, mapping)
# Retain unmapped items
inputs = data.copy()
inputs.update(collected)
return inputs
def _map_output(self, data: Dict,
remapping: Optional[Dict]) -> Dict[str, Any]:
"""KeyMapper outputs from the wrapped transforms by gathering and
renaming data items according to the remapping.
Args:
data (dict): The output of the wrapped pipeline.
remapping (dict, optional): The output key mapping. See the
document of ``mmcv.transforms.wrappers.KeyMapper`` for
details. If ``remapping is None``, no key mapping will be
applied but only remove the special token ``IgnoreKey``.
Returns:
dict: The output with remapped keys.
"""
# Remove ``IgnoreKey``
if remapping is None:
return {k: v for k, v in data.items() if v is not IgnoreKey}
def _map(data, m):
if isinstance(m, dict):
assert isinstance(data, dict)
results = {}
for k_in, k_out in m.items():
assert k_in in data
results.update(_map(data[k_in], k_out))
return results
if isinstance(m, (list, tuple)):
assert isinstance(data, (list, tuple))
assert len(data) == len(m)
results = {}
for m_i, d_i in zip(m, data):
results.update(_map(d_i, m_i))
return results
# ``m is ...`` means the key is marked ignored, in which case the
# inner resuls will not affect the outer results in remapping.
# Another case that will have ``data is IgnoreKey`` is that the
# key is missing in the inputs. In this case, if the inner key is
# created by the wrapped transforms, it will be remapped to the
# corresponding outer key during remapping.
if m is ... or data is IgnoreKey:
return {}
return {m: data}
# Note that unmapped items are not retained, which is different from
# the behavior in _map_input. This is to avoid original data items
# being overwritten by intermediate namesakes
return _map(data, remapping)
def _apply_transforms(self, inputs: Dict) -> Dict:
"""Apply ``self.transforms``.
Note that the special token ``IgnoreKey`` will be invisible to
``self.transforms``, but not removed in this method. It will be
eventually removed in :func:``self._map_output``.
"""
results = inputs.copy()
inputs = {k: v for k, v in inputs.items() if v is not IgnoreKey}
outputs = self.transforms(inputs)
if outputs is None:
raise ValueError(
f'Transforms wrapped by {self.__class__.__name__} should '
'not return None.')
results.update(outputs) # type: ignore
return results
def transform(self, results: Dict) -> Dict:
"""Apply mapping, wrapped transforms and remapping."""
# Apply mapping
inputs = self._map_input(results, self.mapping)
# Apply wrapped transforms
outputs = self._apply_transforms(inputs)
# Apply remapping
outputs = self._map_output(outputs, self.remapping)
results.update(outputs) # type: ignore
return results
def __repr__(self) -> str:
repr_str = self.__class__.__name__
repr_str += f'(transforms = {self.transforms}'
repr_str += f', mapping = {self.mapping}'
repr_str += f', remapping = {self.remapping}'
repr_str += f', auto_remap = {self.auto_remap}'
repr_str += f', allow_nonexist_keys = {self.allow_nonexist_keys})'
return repr_str
@TRANSFORMS.register_module()
class TransformBroadcaster(KeyMapper):
"""A transform wrapper to apply the wrapped transforms to multiple data
items. For example, apply Resize to multiple images.
Args:
transforms (list[dict | callable]): Sequence of transform object or
config dict to be wrapped.
mapping (dict): A dict that defines the input key mapping.
Note that to apply the transforms to multiple data items, the
outer keys of the target items should be remapped as a list with
the standard inner key (The key required by the wrapped transform).
See the following example and the document of
``mmcv.transforms.wrappers.KeyMapper`` for details.
remapping (dict): A dict that defines the output key mapping.
The keys and values have the same meanings and rules as in the
``mapping``. Default: None.
auto_remap (bool, optional): If True, an inverse of the mapping will
be used as the remapping. If auto_remap is not given, it will be
automatically set True if 'remapping' is not given, and vice
versa. Default: None.
allow_nonexist_keys (bool): If False, the outer keys in the mapping
must exist in the input data, or an exception will be raised.
Default: False.
share_random_params (bool): If True, the random transform
(e.g., RandomFlip) will be conducted in a deterministic way and
have the same behavior on all data items. For example, to randomly
flip either both input image and ground-truth image, or none.
Default: False.
.. note::
To apply the transforms to each elements of a list or tuple, instead
of separating data items, you can map the outer key of the target
sequence to the standard inner key. See example 2.
example.
Examples:
>>> # Example 1: Broadcast to enumerated keys, each contains a single
>>> # data element
>>> pipeline = [
>>> dict(type='LoadImageFromFile', key='lq'), # low-quality img
>>> dict(type='LoadImageFromFile', key='gt'), # ground-truth img
>>> # TransformBroadcaster maps multiple outer fields to standard
>>> # the inner field and process them with wrapped transforms
>>> # respectively
>>> dict(type='TransformBroadcaster',
>>> # case 1: from multiple outer fields
>>> mapping={'img': ['lq', 'gt']},
>>> auto_remap=True,
>>> # share_random_param=True means using identical random
>>> # parameters in every processing
>>> share_random_param=True,
>>> transforms=[
>>> dict(type='Crop', crop_size=(384, 384)),
>>> dict(type='Normalize'),
>>> ])
>>> ]
>>> # Example 2: Broadcast to keys that contains data sequences
>>> pipeline = [
>>> dict(type='LoadImageFromFile', key='lq'), # low-quality img
>>> dict(type='LoadImageFromFile', key='gt'), # ground-truth img
>>> # TransformBroadcaster maps multiple outer fields to standard
>>> # the inner field and process them with wrapped transforms
>>> # respectively
>>> dict(type='TransformBroadcaster',
>>> # case 2: from one outer field that contains multiple
>>> # data elements (e.g. a list)
>>> # mapping={'img': 'images'},
>>> auto_remap=True,
>>> share_random_param=True,
>>> transforms=[
>>> dict(type='Crop', crop_size=(384, 384)),
>>> dict(type='Normalize'),
>>> ])
>>> ]
>>> Example 3: Set ignored keys in broadcasting
>>> pipeline = [
>>> dict(type='TransformBroadcaster',
>>> # Broadcast the wrapped transforms to multiple images
>>> # 'lq' and 'gt, but only update 'img_shape' once
>>> mapping={
>>> 'img': ['lq', 'gt'],
>>> 'img_shape': ['img_shape', ...],
>>> },
>>> auto_remap=True,
>>> share_random_params=True,
>>> transforms=[
>>> # `RandomCrop` will modify the field "img",
>>> # and optionally update "img_shape" if it exists
>>> dict(type='RandomCrop'),
>>> ])
>>> ]
"""
def __init__(self,
transforms: List[Union[Dict, Callable[[Dict], Dict]]],
mapping: Optional[Dict] = None,
remapping: Optional[Dict] = None,
auto_remap: Optional[bool] = None,
allow_nonexist_keys: bool = False,
share_random_params: bool = False):
super().__init__(transforms, mapping, remapping, auto_remap,
allow_nonexist_keys)
self.share_random_params = share_random_params
def scatter_sequence(self, data: Dict) -> List[Dict]:
"""Scatter the broadcasting targets to a list of inputs of the wrapped
transforms."""
# infer split number from input
seq_len = 0
key_rep = None
if self.mapping:
keys = self.mapping.keys()
else:
keys = data.keys()
for key in keys:
assert isinstance(data[key], Sequence)
if seq_len:
if len(data[key]) != seq_len:
raise ValueError('Got inconsistent sequence length: '
f'{seq_len} ({key_rep}) vs. '
f'{len(data[key])} ({key})')
else:
seq_len = len(data[key])
key_rep = key
assert seq_len > 0, 'Fail to get the number of broadcasting targets'
scatters = []
for i in range(seq_len): # type: ignore
scatter = data.copy()
for key in keys:
scatter[key] = data[key][i]
scatters.append(scatter)
return scatters
def transform(self, results: Dict):
"""Broadcast wrapped transforms to multiple targets."""
# Apply input remapping
inputs = self._map_input(results, self.mapping)
# Scatter sequential inputs into a list
input_scatters = self.scatter_sequence(inputs)
# Control random parameter sharing with a context manager
if self.share_random_params:
# The context manager :func`:cache_random_params` will let
# cacheable method of the transforms cache their outputs. Thus
# the random parameters will only generated once and shared
# by all data items.
ctx = cache_random_params # type: ignore
else:
ctx = nullcontext # type: ignore
with ctx(self.transforms):
output_scatters = [
self._apply_transforms(_input) for _input in input_scatters
]
# Collate output scatters (list of dict to dict of list)
outputs = {
key: [_output[key] for _output in output_scatters]
for key in output_scatters[0]
}
# Apply remapping
outputs = self._map_output(outputs, self.remapping)
results.update(outputs)
return results
def __repr__(self) -> str:
repr_str = self.__class__.__name__
repr_str += f'(transforms = {self.transforms}'
repr_str += f', mapping = {self.mapping}'
repr_str += f', remapping = {self.remapping}'
repr_str += f', auto_remap = {self.auto_remap}'
repr_str += f', allow_nonexist_keys = {self.allow_nonexist_keys}'
repr_str += f', share_random_params = {self.share_random_params})'
return repr_str
@TRANSFORMS.register_module()
class RandomChoice(BaseTransform):
"""Process data with a randomly chosen transform from given candidates.
Args:
transforms (list[list]): A list of transform candidates, each is a
sequence of transforms.
prob (list[float], optional): The probabilities associated
with each pipeline. The length should be equal to the pipeline
number and the sum should be 1. If not given, a uniform
distribution will be assumed.
Examples:
>>> # config
>>> pipeline = [
>>> dict(type='RandomChoice',
>>> transforms=[
>>> [dict(type='RandomHorizontalFlip')], # subpipeline 1
>>> [dict(type='RandomRotate')], # subpipeline 2
>>> ]
>>> )
>>> ]
"""
def __init__(self,
transforms: List[Union[Transform, List[Transform]]],
prob: Optional[List[float]] = None):
super().__init__()
if prob is not None:
assert mmengine.is_seq_of(prob, float)
assert len(transforms) == len(prob), \
'``transforms`` and ``prob`` must have same lengths. ' \
f'Got {len(transforms)} vs {len(prob)}.'
assert sum(prob) == 1
self.prob = prob
self.transforms = [Compose(transforms) for transforms in transforms]
def __iter__(self):
return iter(self.transforms)
@cache_randomness
def random_pipeline_index(self) -> int:
"""Return a random transform index."""
indices = np.arange(len(self.transforms))
return np.random.choice(indices, p=self.prob)
def transform(self, results: Dict) -> Optional[Dict]:
"""Randomly choose a transform to apply."""
idx = self.random_pipeline_index()
return self.transforms[idx](results)
def __repr__(self) -> str:
repr_str = self.__class__.__name__
repr_str += f'(transforms = {self.transforms}'
repr_str += f'prob = {self.prob})'
return repr_str
@TRANSFORMS.register_module()
class RandomApply(BaseTransform):
"""Apply transforms randomly with a given probability.
Args:
transforms (list[dict | callable]): The transform or transform list
to randomly apply.
prob (float): The probability to apply transforms. Default: 0.5
Examples:
>>> # config
>>> pipeline = [
>>> dict(type='RandomApply',
>>> transforms=[dict(type='HorizontalFlip')],
>>> prob=0.3)
>>> ]
"""
def __init__(self,
transforms: Union[Transform, List[Transform]],
prob: float = 0.5):
super().__init__()
self.prob = prob
self.transforms = Compose(transforms)
def __iter__(self):
return iter(self.transforms)
@cache_randomness
def random_apply(self) -> bool:
"""Return a random bool value indicating whether apply the
transform."""
return np.random.rand() < self.prob
def transform(self, results: Dict) -> Optional[Dict]:
"""Randomly apply the transform."""
if self.random_apply():
return self.transforms(results) # type: ignore
else:
return results
def __repr__(self) -> str:
repr_str = self.__class__.__name__
repr_str += f'(transforms = {self.transforms}'
repr_str += f', prob = {self.prob})'
return repr_str