-
Notifications
You must be signed in to change notification settings - Fork 1
/
assignment.py
143 lines (109 loc) · 4.82 KB
/
assignment.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
import numpy as np
import tensorflow as tf
import argparse
from tensorflow.keras.models import Model
from tensorflow.keras.layers import (
Input,
MaxPool2D,
BatchNormalization,
Activation,
Conv2D,
Dropout,
Concatenate,
AveragePooling2D,
GlobalAveragePooling2D,
Flatten,
Dense
)
"""
IMPLEMENTED BY ADAM, BOWEN, AND KIRAN
Referencing these DenseNet model architectures:
https://github.com/liuzhuang13/DenseNet
https://github.com/taki0112/Densenet-Tensorflow
https://github.com/keras-team/keras-applications/blob/master/keras_applications/densenet.py
"""
from preprocess import get_data
parser = argparse.ArgumentParser(description='ASSIGNMENT')
parser.add_argument('--train-csv-path', type=str, default='MURA-v1.1/train_image_paths.csv',
help='Path to the CSV list of folder sorted images')
parser.add_argument('--test-csv-path', type=str, default='MURA-v1.1/valid_image_paths.csv',
help='Path to the CSV list of folder sorted images')
parser.add_argument('--warp-size', type=tuple, default=(128,128),
help='Shape we warp images to')
parser.add_argument('--growth-k', type=int, default=48,
help='growth of output channel')
parser.add_argument('--drop-rate', type=float, default=.2,
help='for dropout lol')
parser.add_argument('--batch-size', type=int, default=100,
help='literally the batch size')
parser.add_argument('--num-epochs', type=int, default=10,
help='actually for real the number of epochs')
args = parser.parse_args()
def postfix(name):
return lambda x: '{0}_{1}'.format(name, x)
def bn_relu_conv(x0, k, drop_rate, training, name):
pfname = postfix(name)
x1 = BatchNormalization(name = pfname('bn0'))(
x0, training = training)
x1 = Activation('relu', name = pfname('rl0'))(x1)
x1 = Conv2D(4 * k, 1, padding='SAME', use_bias=False, name = pfname('cv0'))(x1)
x1 = Dropout(drop_rate, name = pfname('do0'))(
x1, training = training)
x1 = BatchNormalization(name = pfname('bn1'))(
x1, training = training)
x1 = Activation('relu', name = pfname('rl1'))(x1)
x1 = Conv2D(k, 3, padding='SAME', use_bias=False, name = pfname('cv1'))(x1)
x1 = Dropout(drop_rate, name = pfname('do1'))(
x1, training = training)
return Concatenate(axis=3, name = pfname('cc'))([x0, x1])
def dense_block(x, k, num_bn_relu_conv, drop_rate, training, name):
pfname = postfix(name)
for i in range(num_bn_relu_conv):
x = bn_relu_conv(x, k, drop_rate, training, pfname('brc{}'.format(i)))
return x
def transition_layer(x, k, drop_rate, training, name):
pfname = postfix(name)
x = BatchNormalization(name = pfname('bn'))(x)
x = Activation('relu', name = pfname('rl'))(x)
x = Conv2D(k, 1, padding='SAME', use_bias=False, name = pfname('cv'))(x)
x = Dropout(drop_rate, name = pfname('do'))(
x, training)
x = AveragePooling2D(2, 2, 'VALID')(x)
return x
def DenseNet(shape, k, drop_rate, training):
image_input = Input(shape=shape)
x = Conv2D(2 * k, 7, 2, padding = 'SAME', use_bias = False, name = 'cv0')(
image_input)
x = MaxPool2D(pool_size = 3, strides = 2, padding = 'VALID', name = 'mp0')(x)
x = dense_block(x, k, 6, drop_rate, training, 'db0')
x = transition_layer(x, k, drop_rate, training, 'tl1')
x = dense_block(x, k, 12, drop_rate, training, 'db1')
x = transition_layer(x, k, drop_rate, training, 'tl2')
x = dense_block(x, k, 24, drop_rate, training, 'db2')
x = transition_layer(x, k, drop_rate, training, 'tl3')
x = dense_block(x, k, 16, drop_rate, training, 'db3')
x = BatchNormalization(name = 'bnf')(x, training)
x = Activation('relu', name = 'rl')(x)
x = GlobalAveragePooling2D(name = 'gp')(x)
x = Flatten()(x)
x = Dense(1)(x)
return Model(inputs=image_input, outputs=x, name='densenet')
def main():
# Preprocess data
train_labels, train_images = get_data(args.train_csv_path, args.warp_size)
test_labels, test_images = get_data(args.test_csv_path, args.warp_size)
# Convert to tensors
train_labels = tf.convert_to_tensor(train_labels)
train_images = tf.convert_to_tensor(train_images)
test_labels = tf.convert_to_tensor(test_labels)
test_images = tf.convert_to_tensor(test_images)
# Instantiate, compile, and train
# One channel for our black and white images
input_shape = args.warp_size + (1,)
# TODO: How do we make training=False false later?
model = DenseNet(input_shape, args.growth_k, args.drop_rate, True)
model.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])
model.fit(train_images, train_labels, batch_size=args.batch_size,
epochs=args.num_epochs, validation_data = (test_images, test_labels))
if __name__ == '__main__':
main()