fnet.data package¶
Submodules¶
fnet.data.aicsczidataset module¶
fnet.data.bufferedpatchdataset module¶
-
class
fnet.data.bufferedpatchdataset.
BufferedPatchDataset
(dataset: collections.abc.Sequence, patch_shape: Sequence[int] = (32, 64, 64), buffer_size: int = 1, buffer_switch_interval: int = -1, shuffle_images: bool = True)[source]¶ Bases:
object
Provides patches from items of a dataset.
Parameters: - dataset – Dataset object.
- patch_shape – Shape of patch to be extracted from dataset items.
- buffer_size – Size of buffer.
- buffer_switch_interval – Number of patches provided between buffer item exchanges. Set to -1 to disable exchanges.
- shuffle_images – Set to randomize order of dataset item insertion into buffer.
-
get_batch
(batch_size: int) → Sequence[torch.Tensor][source]¶ Returns a batch of patches.
Parameters: batch_size – Number of patches in batch. Returns: Batch of patches. Return type: Sequence[torch.Tensor]
-
get_buffer_history
() → List[int][source]¶ Returns a list of indices of dataset elements inserted into the buffer.
Returns: Indices of dataset elements. Return type: List[int]
-
get_random_patch
() → List[Union[numpy.ndarray, torch.Tensor]][source]¶ Samples random patch from an item in the buffer.
Let nd be the number of dimensions of the patch. If the item has more dimensions than the patch, then sampling will be from the last nd dimensions of the item.
Returns: Random patch sampled from a dataset item. Return type: List[ArrayLike]
fnet.data.czidataset module¶
fnet.data.czireader module¶
fnet.data.dummychunkdataset module¶
fnet.data.fnetdataset module¶
-
class
fnet.data.fnetdataset.
FnetDataset
(dataframe: Optional[pandas.core.frame.DataFrame] = None, path_csv: Optional[str] = None, transform_signal: Optional[list] = None, transform_target: Optional[list] = None)[source]¶ Bases:
torch.utils.data.dataset.Dataset
Abstract class for fnet datasets.
Parameters: - dataframe – DataFrame where rows are dataset elements. Overrides path_csv.
- path_csv – Path to csv from which to create DataFrame.
- transform_signal – List of transforms to apply to signal image.
- transform_target – List of transforms to apply to target image.
-
get_information
(index) → Union[dict, str][source]¶ Returns information to identify dataset element specified by index.
-
metadata
¶ Returns metadata about the dataset.
fnet.data.tiffdataset module¶
-
class
fnet.data.tiffdataset.
TiffDataset
(col_index: Optional[str] = None, col_signal: str = 'path_signal', col_target: str = 'path_target', col_weight_map: str = 'path_weight_map', augment: bool = False, **kwargs)[source]¶ Bases:
fnet.data.fnetdataset.FnetDataset
Dataset where each row is a signal-target pairing from TIFF files.
- Dataset items will be 2-item or 3-item tuples:
- (signal image, target image) or (signal image, target image, cost map)
Parameters: augment – Set to augment dataset with flips about the x and/or y axis.
fnet.data.tifreader module¶
Module contents¶
-
class
fnet.data.
BufferedPatchDataset
(dataset: collections.abc.Sequence, patch_shape: Sequence[int] = (32, 64, 64), buffer_size: int = 1, buffer_switch_interval: int = -1, shuffle_images: bool = True)[source]¶ Bases:
object
Provides patches from items of a dataset.
Parameters: - dataset – Dataset object.
- patch_shape – Shape of patch to be extracted from dataset items.
- buffer_size – Size of buffer.
- buffer_switch_interval – Number of patches provided between buffer item exchanges. Set to -1 to disable exchanges.
- shuffle_images – Set to randomize order of dataset item insertion into buffer.
-
get_batch
(batch_size: int) → Sequence[torch.Tensor][source]¶ Returns a batch of patches.
Parameters: batch_size – Number of patches in batch. Returns: Batch of patches. Return type: Sequence[torch.Tensor]
-
get_buffer_history
() → List[int][source]¶ Returns a list of indices of dataset elements inserted into the buffer.
Returns: Indices of dataset elements. Return type: List[int]
-
get_random_patch
() → List[Union[numpy.ndarray, torch.Tensor]][source]¶ Samples random patch from an item in the buffer.
Let nd be the number of dimensions of the patch. If the item has more dimensions than the patch, then sampling will be from the last nd dimensions of the item.
Returns: Random patch sampled from a dataset item. Return type: List[ArrayLike]
-
class
fnet.data.
FnetDataset
(dataframe: Optional[pandas.core.frame.DataFrame] = None, path_csv: Optional[str] = None, transform_signal: Optional[list] = None, transform_target: Optional[list] = None)[source]¶ Bases:
torch.utils.data.dataset.Dataset
Abstract class for fnet datasets.
Parameters: - dataframe – DataFrame where rows are dataset elements. Overrides path_csv.
- path_csv – Path to csv from which to create DataFrame.
- transform_signal – List of transforms to apply to signal image.
- transform_target – List of transforms to apply to target image.
-
get_information
(index) → Union[dict, str][source]¶ Returns information to identify dataset element specified by index.
-
metadata
¶ Returns metadata about the dataset.
-
class
fnet.data.
TiffDataset
(col_index: Optional[str] = None, col_signal: str = 'path_signal', col_target: str = 'path_target', col_weight_map: str = 'path_weight_map', augment: bool = False, **kwargs)[source]¶ Bases:
fnet.data.fnetdataset.FnetDataset
Dataset where each row is a signal-target pairing from TIFF files.
- Dataset items will be 2-item or 3-item tuples:
- (signal image, target image) or (signal image, target image, cost map)
Parameters: augment – Set to augment dataset with flips about the x and/or y axis.
-
class
fnet.data.
MultiChTiffDataset
(dataframe: pandas.core.frame.DataFrame = None, path_csv: str = None, transform_signal=None, transform_target=None)[source]¶ Bases:
fnet.data.fnetdataset.FnetDataset
Dataset for multi-channel tiff files.
Currently assumes that images are loaded in ZCXY format