/
model.py
151 lines (124 loc) · 7.42 KB
/
model.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
#!/usr/bin/env python
from __future__ import print_function
import sys
import gc
import os
import time
import yaml
import traceback
import numpy as np
import theano
import theano.tensor as T
from lasagne.layers.normalization import batch_norm
from lasagne.regularization import regularize_layer_params, l2
import lasagne
from lasagne.layers import Conv2DLayer as ConvLayer
#from lasagne.layers.dnn import Conv2DDNNLayer as ConvLayer
from lasagne.layers import ElemwiseSumLayer
from lasagne.layers import InputLayer
from lasagne.layers import DenseLayer
from lasagne.layers import GlobalPoolLayer
from lasagne.layers import PadLayer
from lasagne.layers import ExpressionLayer
from lasagne.layers import NonlinearityLayer
from lasagne.nonlinearities import softmax, rectify
from lasagne.layers import batch_norm
def jaccard_distance(rounded_prediction, rounded_truth):
n = T.cast(T.sum(rounded_prediction * rounded_truth), 'float32')
a = T.cast(T.sum(rounded_prediction), 'float32')
b = T.cast(T.sum(rounded_truth), 'float32')
z = a + b - n
return T.switch(T.eq(a + b, 0.), 0.0, 1.0 - ((n + 1.) / (z + 1.)))
def inverse_dice(rounded_prediction, rounded_truth):
#rounded_truth = T.iround(rounded_truth)
#n = T.cast(T.sum(rounded_prediction & rounded_truth), 'float32')
#a = T.cast(T.sum(rounded_prediction), 'float32')
#b = T.cast(T.sum(rounded_truth), 'float32')
return -2. * (T.sum(rounded_prediction*rounded_truth)+10.)/(T.sum(rounded_prediction) + T.sum(rounded_truth) + 10.)
#return -2.0 * ( (n + 100.) / (a + b + 100.))
def build_cnn(patch_size, input_var=None):
import pdb; pdb.set_trace()
input_layer = lasagne.layers.InputLayer(shape=(None, 1, patch_size[0], patch_size[1]),
input_var=input_var)
network = batch_norm(lasagne.layers.Conv2DLayer(input_layer, num_filters=32, filter_size=(9,9)))
network = lasagne.layers.MaxPool2DLayer(network, pool_size=2)
network = batch_norm(lasagne.layers.Conv2DLayer(network, num_filters=32, filter_size=(5,5)))
network = lasagne.layers.MaxPool2DLayer(network, pool_size=2)
network = batch_norm(lasagne.layers.Conv2DLayer(network, num_filters=32, filter_size=(5,5)))
network = lasagne.layers.MaxPool2DLayer(network, pool_size=2)
network = batch_norm(lasagne.layers.Conv2DLayer(network, num_filters=32, filter_size=(5,5)))
network = lasagne.layers.MaxPool2DLayer(network, pool_size=2)
network = batch_norm(lasagne.layers.Conv2DLayer(network, num_filters=32, filter_size=(3,3)))
network = batch_norm(lasagne.layers.Conv2DLayer(network, num_filters=32, filter_size=(5,5), pad='full'))
network = lasagne.layers.Upscale2DLayer(network, scale_factor=2)
network = batch_norm(lasagne.layers.Conv2DLayer(network, num_filters=32, filter_size=(3,3), pad='full'))
network = lasagne.layers.Upscale2DLayer(network, scale_factor=2)
network = batch_norm(lasagne.layers.Conv2DLayer(network, num_filters=32, filter_size=(3,3), pad='full'))
network = lasagne.layers.Upscale2DLayer(network, scale_factor=2)
network = batch_norm(lasagne.layers.Conv2DLayer(network, num_filters=32, filter_size=(3,3), pad='full'))
network = lasagne.layers.Upscale2DLayer(network, scale_factor=2)
network = batch_norm(lasagne.layers.Conv2DLayer(network, num_filters=1, filter_size=(5,5), pad='full', nonlinearity=lasagne.nonlinearities.sigmoid))
return network
def build_cnn_u(patch_size, input_var=None, number_layers=4):
network = lasagne.layers.InputLayer(shape=(None, 1, patch_size[0], patch_size[1]),
input_var=input_var)
layers = []
for i in range(0, number_layers):
network = batch_norm(lasagne.layers.Conv2DLayer(network, num_filters=32*(2**i), filter_size=(3,3), pad='same'))
network = batch_norm(lasagne.layers.Conv2DLayer(network, num_filters=32*(2**i), filter_size=(3,3), pad='same'))
# append layers to add shortcuts
layers.append(network)
network = lasagne.layers.MaxPool2DLayer(network, pool_size=(2,2))
# layer without upscale
network = batch_norm(lasagne.layers.Conv2DLayer(network, num_filters=32*(2**(number_layers)), filter_size=(3,3), pad='same'))
network = batch_norm(lasagne.layers.Conv2DLayer(network, num_filters=32*(2**(number_layers)), filter_size=(3,3), pad='same'))
for i in range(number_layers, 0, -1):
network = lasagne.layers.Upscale2DLayer(network, scale_factor=(2,2))
network = lasagne.layers.ConcatLayer([network, layers.pop()], axis=1)
network = batch_norm(lasagne.layers.Conv2DLayer(network, num_filters=32*(2**(i-1)), filter_size=(3,3), pad='same'))
network = batch_norm(lasagne.layers.Conv2DLayer(network, num_filters=32*(2**(i-1)), filter_size=(3,3), pad='same'))
network = batch_norm(lasagne.layers.Conv2DLayer(network, num_filters=1, filter_size=(1,1), nonlinearity=lasagne.nonlinearities.sigmoid))
return network
def compose_prediction_functions(patch_size, scope="default"):
input_var = T.ftensor4(scope + 'inputs')
target_var = T.ftensor4(scope + 'targets')
network = build_cnn(patch_size, input_var)
test_prediction = lasagne.layers.get_output(network, deterministic=True)
#test_prediction = T.iround(test_prediction)
val_fn = theano.function([input_var], [test_prediction])
return network, val_fn
def compose_functions(patch_size, scope="default"):
# Prepare Theano variables for inputs and targets
input_var = T.ftensor4(scope + 'inputs')
#TODO: change labels to just be 1
target_var = T.ftensor4(scope + 'targets')
network = build_cnn(patch_size, input_var)
# Create a loss expression for training, i.e., a scalar objective we want
# to minimize (for our multi-class problem, it is the cross-entropy loss):
prediction = lasagne.layers.get_output(network)
#prediction = prediction.reshape((1, prediction.shape[0]))
# TODO: change this back...
loss = inverse_dice(prediction, target_var)
#loss = lasagne.objectives.squared_error(prediction, target_var).mean()#jaccard_distance(prediction, target_var)
#loss += regularize_layer_params(network, l2) * .0001
# loss = loss.mean()
# We could add some weight decay as well here, see lasagne.regularization.
# Create update expressions for training, i.e., how to modify the
# parameters at each training step. Here, we'll use Stochastic Gradient
# Descent (SGD) with Nesterov momentum, but Lasagne offers plenty more.
params = lasagne.layers.get_all_params(network, trainable=True)
lr = 0.1
sh_lr = theano.shared(lasagne.utils.floatX(lr))
updates = lasagne.updates.nesterov_momentum(loss, params, learning_rate=sh_lr)
# Create a loss expression for validation/testing. The crucial difference
# here is that we do a deterministic forward pass through the network,
# disabling dropout layers.
test_prediction = lasagne.layers.get_output(network, deterministic=True)
#test_loss = inverse_dice(test_prediction, target_var)
test_loss = jaccard_distance(test_prediction, target_var)
# Compile a function performing a training step on a mini-batch (by giving
# the updates dictionary) and returning the corresponding training loss:
train_fn = theano.function([input_var, target_var], loss, updates=updates)
# Compile a second function computing the validation loss and accuracy:
val_fn = theano.function([input_var, target_var], [test_loss, test_prediction])
return network, train_fn, val_fn, sh_lr