Coverage for /scratch/builds/bob/bob.med.tb/miniconda/conda-bld/bob.med.tb_1637571489937/_test_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placeho/lib/python3.8/site-packages/bob/med/tb/data/transforms.py: 100%

1#!/usr/bin/env python 

2# -*- coding: utf-8 -*- 

3 

4"""Image transformations for our pipelines 

5 

6Differences between methods here and those from 

7:py:mod:`torchvision.transforms` is that these support multiple simultaneous 

8image inputs, which are required to feed segmentation networks (e.g. image and 

9labels or masks). We also take care of data augmentations, in which random 

10flipping and rotation needs to be applied across all input images, but color 

11jittering, for example, only on the input image. 

12""" 

13 

14import random 

15 

16import numpy 

17import PIL.Image 

18import torchvision.transforms 

19import torchvision.transforms.functional 

20from scipy.ndimage.filters import gaussian_filter 

21from scipy.ndimage.interpolation import map_coordinates 

22 

23class SingleAutoLevel16to8: 

24 """Converts a 16-bit image to 8-bit representation using "auto-level" 

25 

26 This transform assumes that the input image is gray-scaled. 

27 

28 To auto-level, we calculate the maximum and the minimum of the image, and 

29 consider such a range should be mapped to the [0,255] range of the 

30 destination image. 

31 

32 """ 

33 

34 def __call__(self, img): 

35 imin, imax = img.getextrema() 

36 irange = imax - imin 

37 return PIL.Image.fromarray( 

38 numpy.round( 

39 255.0 * (numpy.array(img).astype(float) - imin) / irange 

40 ).astype("uint8"), 

41 ).convert("L") 

42 

43 

44class RemoveBlackBorders: 

45 """Remove black borders of CXR""" 

46 def __init__(self, threshold=0): 

47 self.threshold = threshold 

48 

49 def __call__(self, img): 

50 img = numpy.asarray(img) 

51 mask = numpy.asarray(img) > self.threshold 

52 return PIL.Image.fromarray( 

53 img[numpy.ix_(mask.any(1), mask.any(0))] 

54 ) 

55 

56class ElasticDeformation: 

57 """Elastic deformation of 2D image slightly adapted from [SIMARD-2003]_. 

58 .. [SIMARD-2003] Simard, Steinkraus and Platt, "Best Practices for 

59 Convolutional Neural Networks applied to Visual Document Analysis", in 

60 Proc. of the International Conference on Document Analysis and 

61 Recognition, 2003. 

62 Source: https://gist.github.com/oeway/2e3b989e0343f0884388ed7ed82eb3b0 

63 """ 

64 def __init__(self, alpha=1000, sigma=30, spline_order=1, mode='nearest', random_state=numpy.random, p=1): 

65 self.alpha = alpha 

66 self.sigma = sigma 

67 self.spline_order = spline_order 

68 self.mode = mode 

69 self.random_state = random_state 

70 self.p = p 

71 

72 def __call__(self, img): 

73 

74 if random.random() < self.p: 

75 

76 img = numpy.asarray(img) 

77 

78 assert img.ndim == 2 

79 

80 shape = img.shape 

81 

82 dx = gaussian_filter((self.random_state.rand(*shape) * 2 - 1), 

83 self.sigma, mode="constant", cval=0) * self.alpha 

84 dy = gaussian_filter((self.random_state.rand(*shape) * 2 - 1), 

85 self.sigma, mode="constant", cval=0) * self.alpha 

86 

87 x, y = numpy.meshgrid(numpy.arange(shape[0]), numpy.arange(shape[1]), indexing='ij') 

88 indices = [numpy.reshape(x + dx, (-1, 1)), numpy.reshape(y + dy, (-1, 1))] 

89 result = numpy.empty_like(img) 

90 result[:, :] = map_coordinates( 

91 img[:, :], indices, order=self.spline_order, mode=self.mode).reshape(shape) 

92 return PIL.Image.fromarray(result) 

93 else: 

94 return img