Upload prediction.py
Browse files- prediction.py +163 -163
prediction.py
CHANGED
@@ -1,164 +1,164 @@
|
|
1 |
-
# example of using saved cycleGAN models for image translation
|
2 |
-
#based on https://machinelearningmastery.com/cyclegan-tutorial-with-keras/
|
3 |
-
from keras.models import load_model
|
4 |
-
import numpy as np
|
5 |
-
import tensorflow_addons as tfa
|
6 |
-
from scipy.ndimage import zoom
|
7 |
-
from tqdm import tqdm
|
8 |
-
import warnings
|
9 |
-
warnings.filterwarnings("ignore")
|
10 |
-
from huggingface_hub import hf_hub_download
|
11 |
-
from skimage.morphology import binary_erosion, binary_dilation
|
12 |
-
from skimage import draw
|
13 |
-
|
14 |
-
|
15 |
-
def predict_mask(image, dim_x, dim_y, dim_z, _resize=True, norm_=True, mode_='test', patch_size=(64,128,128,1), _step=64, _step_z=32, _patch_size_z=64):
|
16 |
-
|
17 |
-
cust={'InstanceNormalization': tfa.layers.InstanceNormalization}
|
18 |
-
#load the model
|
19 |
-
# Download the model from Hugging Face Model Hub
|
20 |
-
model_dir = hf_hub_download(repo_id="Hemaxi/3DCycleGAN", filename="CycleGANVesselSegmentation.h5")
|
21 |
-
model_BtoA = load_model(model_dir, cust)
|
22 |
-
|
23 |
-
print('Mode: {}'.format(mode_))
|
24 |
-
|
25 |
-
_patch_size = patch_size[1]
|
26 |
-
_nbslices = patch_size[0]
|
27 |
-
|
28 |
-
perceqmin = 1
|
29 |
-
perceqmax = 99
|
30 |
-
|
31 |
-
image = ((image/(np.max(image)))*255).astype('uint8')
|
32 |
-
|
33 |
-
print('Image Shape: {}'.format(image.shape))
|
34 |
-
print('----------------------------------------')
|
35 |
-
|
36 |
-
initial_image_x = np.shape(image)[0]
|
37 |
-
initial_image_y = np.shape(image)[1]
|
38 |
-
initial_image_z = np.shape(image)[2]
|
39 |
-
|
40 |
-
#percentile equalization
|
41 |
-
if norm_:
|
42 |
-
minval = np.percentile(image, perceqmin)
|
43 |
-
maxval = np.percentile(image, perceqmax)
|
44 |
-
image = np.clip(image, minval, maxval)
|
45 |
-
image = (((image - minval) / (maxval - minval)) * 255).astype('uint8')
|
46 |
-
|
47 |
-
if _resize:
|
48 |
-
image = zoom(image, (dim_x/0.333, dim_y/0.333, dim_z/0.5), order=3, mode='nearest')
|
49 |
-
image = ((image/np.max(image))*255.0).astype('uint8')
|
50 |
-
|
51 |
-
|
52 |
-
#image size
|
53 |
-
size_y = np.shape(image)[0]
|
54 |
-
size_x = np.shape(image)[1]
|
55 |
-
size_depth = np.shape(image)[2]
|
56 |
-
aux_sizes_or = [size_y, size_x, size_depth]
|
57 |
-
|
58 |
-
|
59 |
-
#patch size
|
60 |
-
new_size_y = int((size_y/_patch_size) + 1) * _patch_size
|
61 |
-
new_size_x = int((size_x/_patch_size) + 1) * _patch_size
|
62 |
-
new_size_z = int((size_depth/_patch_size_z) + 1) * _patch_size_z
|
63 |
-
aux_sizes = [new_size_y, new_size_x, new_size_z]
|
64 |
-
|
65 |
-
## zero padding
|
66 |
-
aux_img = np.random.randint(1,50,(aux_sizes[0], aux_sizes[1], aux_sizes[2]))
|
67 |
-
aux_img[0:aux_sizes_or[0], 0:aux_sizes_or[1],0:aux_sizes_or[2]] = image
|
68 |
-
image = aux_img
|
69 |
-
del aux_img
|
70 |
-
|
71 |
-
final_mask_foreground = np.zeros((np.shape(image)[0], np.shape(image)[1], np.shape(image)[2]))
|
72 |
-
final_mask_background = np.zeros((np.shape(image)[0], np.shape(image)[1], np.shape(image)[2]))
|
73 |
-
final_mask_background = final_mask_background.astype('uint8')
|
74 |
-
final_mask_foreground = final_mask_foreground.astype('uint8')
|
75 |
-
|
76 |
-
|
77 |
-
total_iterations = int(image.shape[0]/_patch_size)
|
78 |
-
|
79 |
-
with tqdm(total=total_iterations) as pbar:
|
80 |
-
i=0
|
81 |
-
while i+_patch_size<=image.shape[0]:
|
82 |
-
j=0
|
83 |
-
while j+_patch_size<=image.shape[1]:
|
84 |
-
k=0
|
85 |
-
while k+_patch_size_z<=image.shape[2]:
|
86 |
-
|
87 |
-
B_real = np.zeros((1,_nbslices,_patch_size,_patch_size,1),dtype='float32')
|
88 |
-
_slice = image[i:i+_patch_size, j:j+_patch_size, k:k+_patch_size_z]
|
89 |
-
|
90 |
-
_slice = _slice.transpose(2,0,1)
|
91 |
-
_slice = np.expand_dims(_slice, axis=-1)
|
92 |
-
|
93 |
-
B_real[0,:]=(_slice-127.5) /127.5
|
94 |
-
|
95 |
-
A_generated = model_BtoA.predict(B_real)
|
96 |
-
|
97 |
-
A_generated = (A_generated + 1)/2 #from [-1,1] to [0,1]
|
98 |
-
|
99 |
-
A_generated = A_generated[0,:,:,:,0]
|
100 |
-
A_generated = A_generated.transpose(1,2,0)
|
101 |
-
|
102 |
-
#print(np.unique(A_generated))
|
103 |
-
A_generated = (A_generated>0.5)*1
|
104 |
-
|
105 |
-
A_generated = A_generated.astype('uint8')
|
106 |
-
|
107 |
-
final_mask_foreground[i:i+_patch_size, j:j+_patch_size, k:k+_patch_size_z] = final_mask_foreground[i:i+_patch_size, j:j+_patch_size, k:k+_patch_size_z] + A_generated
|
108 |
-
final_mask_background[i:i+_patch_size, j:j+_patch_size, k:k+_patch_size_z] = final_mask_background[i:i+_patch_size, j:j+_patch_size, k:k+_patch_size_z] + (1-A_generated)
|
109 |
-
|
110 |
-
k=k+_step_z
|
111 |
-
j=j+_step
|
112 |
-
i=i+_step
|
113 |
-
pbar.update(1)
|
114 |
-
|
115 |
-
|
116 |
-
del _slice
|
117 |
-
del A_generated
|
118 |
-
del B_real
|
119 |
-
|
120 |
-
final_mask = (final_mask_foreground>=final_mask_background)*1
|
121 |
-
|
122 |
-
image = image[0:aux_sizes_or[0], 0:aux_sizes_or[1],0:size_depth]
|
123 |
-
print('Image Shape: {}'.format(image.shape))
|
124 |
-
print('----------------------------------------')
|
125 |
-
|
126 |
-
final_mask = final_mask[0:aux_sizes_or[0], 0:aux_sizes_or[1],0:aux_sizes_or[2]]
|
127 |
-
|
128 |
-
|
129 |
-
if _resize:
|
130 |
-
final_mask = zoom(final_mask, (0.333/dim_x, 0.333/dim_y, 0.5/dim_z), order=3, mode='nearest')
|
131 |
-
final_mask = (final_mask*255.0).astype('uint8')
|
132 |
-
|
133 |
-
final_size_x = np.shape(final_mask)[0]
|
134 |
-
final_size_y = np.shape(final_mask)[1]
|
135 |
-
final_size_z = np.shape(final_mask)[2]
|
136 |
-
|
137 |
-
aux_mask = np.zeros((initial_image_x, initial_image_y, initial_image_z)).astype('uint8')
|
138 |
-
aux_mask[0:min(initial_image_x, final_size_x),0:min(initial_image_y, final_size_y),0:min(initial_image_z, final_size_z)] = final_mask[0:min(initial_image_x, final_size_x),0:min(initial_image_y, final_size_y),0:min(initial_image_z, final_size_z)]
|
139 |
-
|
140 |
-
final_mask = aux_mask.copy()
|
141 |
-
|
142 |
-
|
143 |
-
print('Mask Shape: {}'.format(final_mask.shape))
|
144 |
-
print('----------------------------------------')
|
145 |
-
final_mask = final_mask/np.max(final_mask)
|
146 |
-
final_mask = final_mask*255.0
|
147 |
-
final_mask = final_mask.astype('uint8')
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
|
163 |
-
|
164 |
return closed_mask
|
|
|
1 |
+
# example of using saved cycleGAN models for image translation
|
2 |
+
#based on https://machinelearningmastery.com/cyclegan-tutorial-with-keras/
|
3 |
+
from keras.models import load_model
|
4 |
+
import numpy as np
|
5 |
+
import tensorflow_addons as tfa
|
6 |
+
from scipy.ndimage import zoom
|
7 |
+
from tqdm import tqdm
|
8 |
+
import warnings
|
9 |
+
warnings.filterwarnings("ignore")
|
10 |
+
from huggingface_hub import hf_hub_download
|
11 |
+
from skimage.morphology import binary_erosion, binary_dilation
|
12 |
+
from skimage import draw
|
13 |
+
|
14 |
+
|
15 |
+
def predict_mask(image, dim_x, dim_y, dim_z, _resize=True, norm_=True, mode_='test', patch_size=(64,128,128,1), _step=64, _step_z=32, _patch_size_z=64):
|
16 |
+
|
17 |
+
cust={'InstanceNormalization': tfa.layers.InstanceNormalization}
|
18 |
+
#load the model
|
19 |
+
# Download the model from Hugging Face Model Hub
|
20 |
+
model_dir = hf_hub_download(repo_id="Hemaxi/3DCycleGAN", filename="CycleGANVesselSegmentation.h5")
|
21 |
+
model_BtoA = load_model(model_dir, cust)
|
22 |
+
|
23 |
+
print('Mode: {}'.format(mode_))
|
24 |
+
|
25 |
+
_patch_size = patch_size[1]
|
26 |
+
_nbslices = patch_size[0]
|
27 |
+
|
28 |
+
perceqmin = 1
|
29 |
+
perceqmax = 99
|
30 |
+
|
31 |
+
image = ((image/(np.max(image)))*255).astype('uint8')
|
32 |
+
|
33 |
+
print('Image Shape: {}'.format(image.shape))
|
34 |
+
print('----------------------------------------')
|
35 |
+
|
36 |
+
initial_image_x = np.shape(image)[0]
|
37 |
+
initial_image_y = np.shape(image)[1]
|
38 |
+
initial_image_z = np.shape(image)[2]
|
39 |
+
|
40 |
+
#percentile equalization
|
41 |
+
if norm_:
|
42 |
+
minval = np.percentile(image, perceqmin)
|
43 |
+
maxval = np.percentile(image, perceqmax)
|
44 |
+
image = np.clip(image, minval, maxval)
|
45 |
+
image = (((image - minval) / (maxval - minval)) * 255).astype('uint8')
|
46 |
+
|
47 |
+
if _resize:
|
48 |
+
image = zoom(image, (dim_x/0.333, dim_y/0.333, dim_z/0.5), order=3, mode='nearest')
|
49 |
+
image = ((image/np.max(image))*255.0).astype('uint8')
|
50 |
+
|
51 |
+
|
52 |
+
#image size
|
53 |
+
size_y = np.shape(image)[0]
|
54 |
+
size_x = np.shape(image)[1]
|
55 |
+
size_depth = np.shape(image)[2]
|
56 |
+
aux_sizes_or = [size_y, size_x, size_depth]
|
57 |
+
|
58 |
+
|
59 |
+
#patch size
|
60 |
+
new_size_y = int((size_y/_patch_size) + 1) * _patch_size
|
61 |
+
new_size_x = int((size_x/_patch_size) + 1) * _patch_size
|
62 |
+
new_size_z = int((size_depth/_patch_size_z) + 1) * _patch_size_z
|
63 |
+
aux_sizes = [new_size_y, new_size_x, new_size_z]
|
64 |
+
|
65 |
+
## zero padding
|
66 |
+
aux_img = np.random.randint(1,50,(aux_sizes[0], aux_sizes[1], aux_sizes[2]))
|
67 |
+
aux_img[0:aux_sizes_or[0], 0:aux_sizes_or[1],0:aux_sizes_or[2]] = image
|
68 |
+
image = aux_img
|
69 |
+
del aux_img
|
70 |
+
|
71 |
+
final_mask_foreground = np.zeros((np.shape(image)[0], np.shape(image)[1], np.shape(image)[2]))
|
72 |
+
final_mask_background = np.zeros((np.shape(image)[0], np.shape(image)[1], np.shape(image)[2]))
|
73 |
+
final_mask_background = final_mask_background.astype('uint8')
|
74 |
+
final_mask_foreground = final_mask_foreground.astype('uint8')
|
75 |
+
|
76 |
+
|
77 |
+
total_iterations = int(image.shape[0]/_patch_size)
|
78 |
+
|
79 |
+
with tqdm(total=total_iterations) as pbar:
|
80 |
+
i=0
|
81 |
+
while i+_patch_size<=image.shape[0]:
|
82 |
+
j=0
|
83 |
+
while j+_patch_size<=image.shape[1]:
|
84 |
+
k=0
|
85 |
+
while k+_patch_size_z<=image.shape[2]:
|
86 |
+
|
87 |
+
B_real = np.zeros((1,_nbslices,_patch_size,_patch_size,1),dtype='float32')
|
88 |
+
_slice = image[i:i+_patch_size, j:j+_patch_size, k:k+_patch_size_z]
|
89 |
+
|
90 |
+
_slice = _slice.transpose(2,0,1)
|
91 |
+
_slice = np.expand_dims(_slice, axis=-1)
|
92 |
+
|
93 |
+
B_real[0,:]=(_slice-127.5) /127.5
|
94 |
+
|
95 |
+
A_generated = model_BtoA.predict(B_real)
|
96 |
+
|
97 |
+
A_generated = (A_generated + 1)/2 #from [-1,1] to [0,1]
|
98 |
+
|
99 |
+
A_generated = A_generated[0,:,:,:,0]
|
100 |
+
A_generated = A_generated.transpose(1,2,0)
|
101 |
+
|
102 |
+
#print(np.unique(A_generated))
|
103 |
+
A_generated = (A_generated>0.5)*1
|
104 |
+
|
105 |
+
A_generated = A_generated.astype('uint8')
|
106 |
+
|
107 |
+
final_mask_foreground[i:i+_patch_size, j:j+_patch_size, k:k+_patch_size_z] = final_mask_foreground[i:i+_patch_size, j:j+_patch_size, k:k+_patch_size_z] + A_generated
|
108 |
+
final_mask_background[i:i+_patch_size, j:j+_patch_size, k:k+_patch_size_z] = final_mask_background[i:i+_patch_size, j:j+_patch_size, k:k+_patch_size_z] + (1-A_generated)
|
109 |
+
|
110 |
+
k=k+_step_z
|
111 |
+
j=j+_step
|
112 |
+
i=i+_step
|
113 |
+
pbar.update(1)
|
114 |
+
|
115 |
+
|
116 |
+
del _slice
|
117 |
+
del A_generated
|
118 |
+
del B_real
|
119 |
+
|
120 |
+
final_mask = (final_mask_foreground>=final_mask_background)*1
|
121 |
+
|
122 |
+
image = image[0:aux_sizes_or[0], 0:aux_sizes_or[1],0:size_depth]
|
123 |
+
print('Image Shape: {}'.format(image.shape))
|
124 |
+
print('----------------------------------------')
|
125 |
+
|
126 |
+
final_mask = final_mask[0:aux_sizes_or[0], 0:aux_sizes_or[1],0:aux_sizes_or[2]]
|
127 |
+
|
128 |
+
|
129 |
+
if _resize:
|
130 |
+
final_mask = zoom(final_mask, (0.333/dim_x, 0.333/dim_y, 0.5/dim_z), order=3, mode='nearest')
|
131 |
+
final_mask = (final_mask*255.0).astype('uint8')
|
132 |
+
|
133 |
+
final_size_x = np.shape(final_mask)[0]
|
134 |
+
final_size_y = np.shape(final_mask)[1]
|
135 |
+
final_size_z = np.shape(final_mask)[2]
|
136 |
+
|
137 |
+
aux_mask = np.zeros((initial_image_x, initial_image_y, initial_image_z)).astype('uint8')
|
138 |
+
aux_mask[0:min(initial_image_x, final_size_x),0:min(initial_image_y, final_size_y),0:min(initial_image_z, final_size_z)] = final_mask[0:min(initial_image_x, final_size_x),0:min(initial_image_y, final_size_y),0:min(initial_image_z, final_size_z)]
|
139 |
+
|
140 |
+
final_mask = aux_mask.copy()
|
141 |
+
|
142 |
+
|
143 |
+
print('Mask Shape: {}'.format(final_mask.shape))
|
144 |
+
print('----------------------------------------')
|
145 |
+
final_mask = final_mask/np.max(final_mask)
|
146 |
+
final_mask = final_mask*255.0
|
147 |
+
final_mask = final_mask.astype('uint8')
|
148 |
+
|
149 |
+
|
150 |
+
#closing operation to fill small holes
|
151 |
+
mask = final_mask
|
152 |
+
mask[mask!=0] = 1
|
153 |
+
mask = mask.astype('uint8')
|
154 |
+
|
155 |
+
ellipsoid = draw.ellipsoid(9,9,3, spacing=(1,1,1), levelset=False)
|
156 |
+
ellipsoid = ellipsoid.astype('uint8')
|
157 |
+
ellipsoid = ellipsoid[1:-1,1:-1,1:-1]
|
158 |
+
|
159 |
+
#perform closing operation on the mask
|
160 |
+
dil = binary_dilation(mask, ellipsoid)
|
161 |
+
closed_mask = binary_erosion(dil, ellipsoid)
|
162 |
+
closed_mask = (closed_mask*255.0).astype('uint8')
|
163 |
+
|
164 |
return closed_mask
|