|
a |
|
b/model_definition/baseline.py |
|
|
1 |
import tensorflow as tf |
|
|
2 |
import config |
|
|
3 |
|
|
|
4 |
from model_utils import calculate_conv_output_size |
|
|
5 |
|
|
|
6 |
|
|
|
7 |
n_x = config.IMAGE_PXL_SIZE_X |
|
|
8 |
n_y = config.IMAGE_PXL_SIZE_Y |
|
|
9 |
n_z = config.SLICES |
|
|
10 |
|
|
|
11 |
# This handles padding in both convolution and pooling layers |
|
|
12 |
strides = [[1, 1, 1], |
|
|
13 |
[2, 4, 4], |
|
|
14 |
[1, 1, 1], |
|
|
15 |
[2, 2, 2], |
|
|
16 |
[1, 1, 1], |
|
|
17 |
[2, 2, 2]] |
|
|
18 |
|
|
|
19 |
filters = [[3, 5, 5], |
|
|
20 |
[3, 5, 5], |
|
|
21 |
[3, 3, 3], |
|
|
22 |
[3, 3, 3], |
|
|
23 |
[3, 3, 3], |
|
|
24 |
[3, 3, 3]] |
|
|
25 |
|
|
|
26 |
padding_types = ['VALID'] * 6 |
|
|
27 |
|
|
|
28 |
|
|
|
29 |
baseline_config = { |
|
|
30 |
'weights': [ |
|
|
31 |
# Convolution layers |
|
|
32 |
('wc1', tf.truncated_normal([3, 5, 5, config.NUM_CHANNELS, 16], stddev=0.01)), |
|
|
33 |
('wc2', tf.truncated_normal([3, 3, 3, 16, 32], stddev=0.01)), |
|
|
34 |
('wc3', tf.truncated_normal([3, 3, 3, 32, 32], stddev=0.01)), |
|
|
35 |
# Fully connected layers |
|
|
36 |
('wd1', tf.truncated_normal([calculate_conv_output_size(n_x, n_y, n_z, |
|
|
37 |
strides, |
|
|
38 |
filters, |
|
|
39 |
padding_types, |
|
|
40 |
32), |
|
|
41 |
100], stddev=0.01)), |
|
|
42 |
('wout', tf.truncated_normal([100, config.N_CLASSES], stddev=0.01)) |
|
|
43 |
], |
|
|
44 |
'biases': [ |
|
|
45 |
# Convolution layers |
|
|
46 |
('bc1', tf.zeros([16])), |
|
|
47 |
('bc2', tf.constant(1.0, shape=[32])), |
|
|
48 |
('bc3', tf.zeros([32])), |
|
|
49 |
# Fully connected layers |
|
|
50 |
('bd1', tf.constant(1.0, shape=[100])), |
|
|
51 |
('bout', tf.constant(1.0, shape=[config.N_CLASSES])) |
|
|
52 |
], |
|
|
53 |
'pool_strides': [ |
|
|
54 |
[1, 2, 4, 4, 1], |
|
|
55 |
[1, 2, 2, 2, 1], |
|
|
56 |
[1, 2, 2, 2, 1], |
|
|
57 |
], |
|
|
58 |
'pool_windows': [ |
|
|
59 |
[1, 3, 5, 5, 1], |
|
|
60 |
[1, 3, 3, 3, 1], |
|
|
61 |
[1, 3, 3, 3, 1], |
|
|
62 |
], |
|
|
63 |
'strides': [ |
|
|
64 |
[1, 1, 1, 1, 1], |
|
|
65 |
[1, 1, 1, 1, 1], |
|
|
66 |
[1, 1, 1, 1, 1], |
|
|
67 |
] |
|
|
68 |
} |