# @Copyright: CEA-LIST/DIASI/SIALV/ (2023- )
# @Author: CEA-LIST/DIASI/SIALV/ <julien.denize@cea.fr>
# @License: CECILL-C
from __future__ import annotations
from typing import Any, List, Mapping, Optional, Sequence, Tuple, Union
import torch
from torch import Tensor
from torch.utils._pytree import tree_flatten
from ._batch_concatenated_ta_tensor import _BatchConcatenatedTATensor
from ._mask import Mask
_CHECK_ATTRS = [
"requires_grad",
"device",
"dtype",
]
[docs]
def convert_masks_to_batch_masks(
masks: Sequence[Mask],
) -> BatchMasks:
"""Convert a sequence of :class:`~torchaug.ta_tensors.Mask` objects to a
:class:`~torchaug.ta_tensors.BatchMasks` object.
"""
for mask in masks:
if not mask.shape[-2:] == masks[0].shape[-2:]:
raise ValueError("All masks must have the same size.")
for attr in _CHECK_ATTRS:
if getattr(mask, attr) != getattr(masks[0], attr):
raise ValueError(f"All masks must have the same {attr} attribute.")
masks = [mask if mask.ndim > 2 else mask.unsqueeze(0) for mask in masks]
masks_data = torch.cat(masks)
samples_ranges = []
sum_masks = 0
for mask in masks:
samples_ranges.append((sum_masks, sum_masks + mask.shape[0]))
sum_masks += mask.shape[0]
batch_masks = BatchMasks(
masks_data,
samples_ranges=samples_ranges,
)
return batch_masks
[docs]
def convert_batch_masks_to_masks(
batch_masks: BatchMasks,
) -> List[Mask]:
"""Convert :class:`~torchaug.ta_tensors.BatchMasks` object to a list of
:class:`~torchaug.ta_tensors.Mask` objects.
"""
samples_ranges = batch_masks.samples_ranges
list_masks = [
Mask(
batch_masks[idx_start:idx_stop],
)
for (idx_start, idx_stop) in samples_ranges
]
return list_masks
[docs]
class BatchMasks(_BatchConcatenatedTATensor):
""":class:`torch.Tensor` subclass for batch of segmentation and detection masks.
Args:
data: Any data that can be turned into a tensor with :func:`torch.as_tensor`.
dtype: Desired data type. If omitted, will be inferred from
``data``.
samples_ranges: Each element is the range of the indices of the masks for each sample.
device: Desired device. If omitted and ``data`` is a
:class:`torch.Tensor`, the device is taken from it. Otherwise, the mask is constructed on the CPU.
requires_grad: Whether autograd should record operations. If omitted and
``data`` is a :class:`torch.Tensor`, the value is taken from it. Otherwise, defaults to ``False``.
"""
[docs]
@classmethod
def cat(cls, masks_batches: Sequence[BatchMasks]):
"""Concatenates a sequence of :class:`~torchaug.ta_tensors.BatchMasks` along the first dimension.
Args:
masks_batches: A sequence of :class:`~torchaug.ta_tensors.BatchMasks` to concatenate.
Returns:
The concatenated :class:`~torchaug.ta_tensors.BatchMasks`.
"""
for batch_mask in masks_batches:
if not isinstance(batch_mask, BatchMasks):
raise ValueError("All batches must be of type BatchMasks.")
if not batch_mask.shape[-2:] == masks_batches[0].shape[-2:]:
raise ValueError("All batches of masks must have the same size.")
for attr in _CHECK_ATTRS:
if getattr(batch_mask, attr) != getattr(masks_batches[0], attr):
raise ValueError(f"All batches of masks must have the same {attr} attribute.")
samples_ranges = []
sum_masks = 0
for batch_masks in masks_batches:
for idx_start, idx_stop in batch_masks.samples_ranges:
samples_ranges.append((idx_start + sum_masks, idx_stop + sum_masks))
sum_masks += batch_masks.num_data
data = torch.cat([batch_masks.data for batch_masks in masks_batches], 0)
return cls(
data,
samples_ranges=samples_ranges,
)
@classmethod
def _wrap(
cls,
tensor: Tensor,
*,
samples_ranges: List[Tuple[int, int]],
check_dims: bool = True,
) -> BatchMasks: # type: ignore[override]
if check_dims and tensor.ndim < 2:
raise ValueError(f"Expected at least a 2D tensor, got {tensor.ndim}D tensor")
batch_masks = tensor.as_subclass(cls)
batch_masks.samples_ranges = samples_ranges
return batch_masks
def __new__(
cls,
data: Any,
*,
samples_ranges: List[Tuple[int, int]],
dtype: Optional[torch.dtype] = None,
device: Optional[Union[torch.device, str, int]] = None,
requires_grad: Optional[bool] = None,
) -> BatchMasks:
tensor = cls._to_tensor(data, dtype=dtype, device=device, requires_grad=requires_grad)
if tensor.ndim < 3:
raise ValueError
cls._check_samples_ranges(samples_ranges, tensor)
return cls._wrap(tensor, samples_ranges=samples_ranges)
@classmethod
def _wrap_output(
cls,
output: torch.Tensor,
args: Sequence[Any] = (),
kwargs: Optional[Mapping[str, Any]] = None,
) -> BatchMasks:
# If there are BatchMasks instances in the output, their metadata got lost when we called
# super().__torch_function__. We need to restore the metadata somehow, so we choose to take
# the metadata from the first mask in the parameters.
# This should be what we want in most cases. When it's not, it's probably a mis-use anyway, e.g.
# something like mask_N1HW + mask_N2HW; we don't guard against those cases.
flat_params, _ = tree_flatten(args + (tuple(kwargs.values()) if kwargs else ())) # type: ignore[operator]
first_batch_masks_from_args = next(x for x in flat_params if isinstance(x, BatchMasks))
samples_ranges = first_batch_masks_from_args.samples_ranges.copy() # clone the list.
if isinstance(output, torch.Tensor) and not isinstance(output, BatchMasks):
output = BatchMasks._wrap(
output,
samples_ranges=samples_ranges,
check_dims=False,
)
elif isinstance(output, (tuple, list)):
output = type(output)(
BatchMasks._wrap(
part,
samples_ranges=samples_ranges,
check_dims=False,
)
for part in output
)
return output
[docs]
def get_sample(self, idx: int) -> Mask:
"""Get the masks for a sample in the batch.
Args:
idx: The index of the sample to get.
Returns:
The masks for the sample.
"""
masks = self[self.samples_ranges[idx][0] : self.samples_ranges[idx][1]]
return Mask(
masks,
device=self.device,
requires_grad=self.requires_grad,
)
[docs]
def get_chunk(self, chunk_indices: torch.Tensor) -> BatchMasks:
"""Get a chunk of the batch of masks.
Args:
chunk_indices: The indices of the chunk to get.
Returns:
The chunk of the batch of masks.
"""
chunk_samples_ranges = self._get_chunk_samples_ranges_from_chunk_indices(chunk_indices)
data_indices = self._get_data_indices_from_chunk_indices(chunk_indices)
return BatchMasks(
self[data_indices],
samples_ranges=chunk_samples_ranges,
device=self.device,
requires_grad=self.requires_grad,
)
[docs]
def update_chunk_(self, chunk: BatchMasks, chunk_indices: torch.Tensor) -> BatchMasks:
"""Update a chunk of the batch of masks.
Args:
chunk: The chunk update.
chunk_indices: The indices of the chunk to update.
Returns:
The updated batch of masks.
"""
return super().update_chunk_(chunk, chunk_indices)
[docs]
def to_samples(self) -> list[Mask]:
"""Get the tensors."""
return [self.get_sample(i).clone() for i in range(self.batch_size)]
[docs]
@classmethod
def masked_select(cls, masks: BatchMasks, mask: torch.Tensor) -> BatchMasks:
"""Remove masks from the batch of masks.
Args:
masks: The batch of masks to remove masks from.
mask: A boolean mask to keep masks.
Returns:
The updated batch of masks.
"""
old_samples_ranges = masks.samples_ranges
data = masks.data[mask]
neg_mask = (~mask).cpu()
num_delete_per_sample = [
neg_mask[old_samples_ranges[i][0] : old_samples_ranges[i][1]].sum().item()
for i in range(len(old_samples_ranges))
]
new_samples_ranges = [
(
old_samples_ranges[i][0] - sum(num_delete_per_sample[:i]),
old_samples_ranges[i][1] - sum(num_delete_per_sample[: i + 1]),
)
for i in range(len(old_samples_ranges))
]
return cls._wrap(
data,
samples_ranges=new_samples_ranges,
)