niftynet.engine.windows_aggregator_base module

This module is used to cache window-based network outputs, form a image-level output, write the cached the results to hard drive.

class ImageWindowsAggregator(image_reader=None, output_path='.')

Bases: object

Image windows are retrieved and analysed by the tensorflow graph, this windows aggregator receives output window data in numpy array. To access image-level information the reader is needed.

input_image

Get the corresponding input image of these batch data. So that the batch data can be stored correctly in terms of interpolation order, orientation, pixdims.

Returns:an image object from image reader

image_id

Index of the image in the output image list maintained by image reader.

Returns:integer of the position in image list

decode_batch(*args, **kwargs)

The implementation of caching and writing batch output goes here. This function should return False when the location vector is stopping signal, to notify the inference loop to terminate.

Parameters:

• args –
• kwargs –

Returns:

True if more batch data are expected, False otherwise

static crop_batch(window, location, border=None)

This utility function removes two borders along each spatial dim of the output image window data, adjusts window spatial coordinates accordingly.

Parameters:

• window –
• location –
• border –

Returns:

log_inferred(subject_name, filename)

This function writes out a csv of inferred files

Parameters:

• subject_name – subject name corresponding to output
• filename – filename of output

Returns: