Source code for niftynet.engine.windows_aggregator_base

# -*- coding: utf-8 -*-
"""
This module is used to cache window-based network outputs,
form a image-level output,
write the cached the results to hard drive.
"""
from __future__ import absolute_import, print_function, division

import os
import numpy as np
import tensorflow as tf
from niftynet.engine.image_window import N_SPATIAL


[docs]class ImageWindowsAggregator(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. """ def __init__(self, image_reader=None, output_path='.'): self.reader = image_reader self._image_id = None self.postfix = '' self.output_path = os.path.abspath(output_path) if not os.path.exists(self.output_path): os.makedirs(self.output_path) self.inferred_cleared = False @property def input_image(self): """ 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. :return: an image object from image reader """ if self.image_id is not None and self.reader: return self.reader.output_list[self.image_id] return None @property def image_id(self): """ Index of the image in the output image list maintained by image reader. :return: integer of the position in image list """ return self._image_id @image_id.setter def image_id(self, current_id): try: self._image_id = int(current_id) except (ValueError, TypeError): tf.logging.fatal("unknown image id format (should be an integer)")
[docs] def decode_batch(self, *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. :param args: :param kwargs: :return: True if more batch data are expected, False otherwise """ raise NotImplementedError
@staticmethod def _is_stopping_signal(location_vector): return np.any(location_vector < 0)
[docs] @staticmethod def 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. :param window: :param location: :param border: :return: """ if not border: return window, location assert isinstance(border, (list, tuple)), \ "border should be a list or tuple" while len(border) < N_SPATIAL: border = tuple(border) + (border[-1],) border = border[:N_SPATIAL] location = location.astype(np.int) window_shape = window.shape spatial_shape = window_shape[1:-1] n_spatial = len(spatial_shape) for idx in range(n_spatial): location[:, idx + 1] = location[:, idx + 1] + border[idx] location[:, idx + 4] = location[:, idx + 4] - border[idx] if np.any(location < 0): return window, location cropped_shape = np.max(location[:, 4:7] - location[:, 1:4], axis=0) left = np.floor( (spatial_shape - cropped_shape[:n_spatial])/2.0).astype(np.int) if np.any(left < 0): tf.logging.fatal( 'network output window can be ' 'cropped by specifying the border parameter in config file, ' 'but here the output window content size %s is smaller ' 'than the window coordinate size: %s -- ' 'computed by input window size minus border size (%s)' 'not supported by this aggregator. Please try larger border.)', spatial_shape, cropped_shape, border) raise ValueError if n_spatial == 1: window = window[:, left[0]:(left[0] + cropped_shape[0]), np.newaxis, np.newaxis, ...] elif n_spatial == 2: window = window[:, left[0]:(left[0] + cropped_shape[0]), left[1]:(left[1] + cropped_shape[1]), np.newaxis, ...] elif n_spatial == 3: window = window[:, left[0]:(left[0] + cropped_shape[0]), left[1]:(left[1] + cropped_shape[1]), left[2]:(left[2] + cropped_shape[2]), ...] else: tf.logging.fatal( 'unknown output format: shape %s' ' spatial dims are: %s', window_shape, spatial_shape) raise NotImplementedError return window, location
[docs] def log_inferred(self, subject_name, filename): """ This function writes out a csv of inferred files :param subject_name: subject name corresponding to output :param filename: filename of output :return: """ inferred_csv = os.path.join(self.output_path, 'inferred.csv') if not self.inferred_cleared: if os.path.exists(inferred_csv): os.remove(inferred_csv) self.inferred_cleared = True if not os.path.exists(self.output_path): os.makedirs(self.output_path) with open(inferred_csv, 'a+') as csv_file: filename = os.path.join(self.output_path, filename) csv_file.write('{},{}\n'.format(subject_name, filename))