kaishi.image.file_group

Definition for groups of image files.

Module Contents

kaishi.image.file_group.THUMBNAIL_SIZE = [64, 64]

kaishi.image.file_group.MAX_DIM_FOR_SMALL = 224

kaishi.image.file_group.PATCH_SIZE = [64, 64]

kaishi.image.file_group.RESAMPLE_METHOD

class kaishi.image.file_group.ImageFileGroup(source: str, recursive: bool)

Bases: kaishi.core.file_group.FileGroup

Group of image files that inherits from the core file group class.

load_all(self)

Load all files in the directory that this class was initialized with.

build_numpy_batches(self, channels_first: bool = True, batch_size: int = None, image_type: str = 'small_image')

Build a tensor from the entire image corpus (or generate batches if specified).

If a batch size is specified, this acts as a generator of batches and returns a list of file objects to manipulate. Otherwise, a single batch of all images is returned in an array format.

Parameters

• channels_first (bool) – flag indicating channels first (e.g. PyTorch) vs. channels last (e.g. Keras)

• batch_size (int) – size of each batch (default is None, which will return a single batch)

• image_type (str) – choice of “small_image”, “thumbnail”, or “patch”, indicating which version of each image to use

Returns

batch of images (generator if batch size specified)

Return type

numpy.array

save(self, out_dir: str)

Save image data set in the current structure while preserving any changes.

Parameters

out_dir (str) – output directory

run_pipeline(self, verbose: bool = False)

Run the pipeline as configured.

Parameters

verbose (bool) – flag indicating verbosity

report(self)

Run a descriptive report (currently just prints the file report).