-
Notifications
You must be signed in to change notification settings - Fork 0
/
autocompress_6deep_reluout.py
210 lines (174 loc) · 5.54 KB
/
autocompress_6deep_reluout.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
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
#!/bin/python3
import numpy as np
import tensorflow as tf
sess = tf.InteractiveSession()
# Constants
COLOR_MAX = tf.constant(255, dtype = tf.float32)
ITERATIONS = 100
STRIDE = [1, 4, 2, 2, 1]
STRIDE2 = [1, 4, 1, 1, 1]
VOXEL_SHAPE = [64, 240, 320, 3]
SAVEPATH = "Trained_Graphs/TrainedLayers_reluout.ckpt"
first = 64
second = 32
third = 16
reducedDimX = 60
reducedDimY = 80
reducedDimZ = 1
BATCH = 1
def weights (input_filter_size, out_filter_size):
return tf.Variable(
tf.random_normal(
shape = [5, 5, 5, input_filter_size, out_filter_size],
stddev = 0.01,
dtype = tf.float32
)
)
def biases (shape):
return tf.Variable(
tf.zeros(shape = shape, dtype = tf.float32)
)
def clayer (input, weights, biases, stride):
conv = tf.nn.conv3d(input, weights, strides = stride, padding = 'SAME')
return conv + biases
def dlayer(input, output, stride):
#global STRIDE
deconv = tf.nn.conv3d_transpose(input, weights(output[-1], input.get_shape()[-1].value), output_shape = output, strides = stride)
output = deconv + biases(output[-1])
return output
def dense(x, in_features, out_features):
mat = tf.Variable(tf.random_normal(shape = [in_features, out_features], stddev = .01, dtype=tf.float32))
x = tf.cast(x, tf.float32)
mat = tf.cast(mat, tf.float32)
output = tf.matmul(x, mat) + biases(out_features)
return output
def encoder (x):
l1 = tf.nn.elu(clayer(x, weights(3, first), biases(first), STRIDE))
l2 = tf.nn.elu(clayer(l1, weights(first, second),biases(second), STRIDE2))
l3 = tf.nn.relu(clayer(l2, weights(second, third), biases(third), STRIDE))
re = tf.reshape(l3, [BATCH, reducedDimX * reducedDimY * reducedDimZ * third])
return re
def decoder (x):
matrix = (tf.reshape(x, [BATCH, reducedDimZ, reducedDimX, reducedDimY, third]))
l1 = tf.nn.elu(dlayer(matrix, [BATCH, reducedDimZ * 4, reducedDimX * 2, reducedDimY * 2, second], STRIDE))
l2 = tf.nn.elu(dlayer(l1, [BATCH, reducedDimZ * 16, reducedDimX * 2, reducedDimY * 2, first], STRIDE2))
l3 = tf.nn.sigmoid(dlayer(l2, [BATCH, *VOXEL_SHAPE], STRIDE))
print("zep")
return l3
# Some variables used only in process()
iteration_index = 0
# Placeholder for feed
feed = tf.placeholder_with_default(
tf.ones(
shape = [BATCH, *VOXEL_SHAPE],
dtype = tf.float32)
,
shape = [BATCH, *VOXEL_SHAPE],
name = "inputVox"
)
# Create pipeline
# Read in video voxel to tf
dataset = tf.data.Dataset.from_tensor_slices(feed)
batched = dataset.batch(BATCH).repeat()
iterator = batched.make_initializable_iterator()
nextBatch = iterator.get_next()
# Feed into pipeline
latent_vector = encoder(nextBatch)
decompressed = decoder(latent_vector)
loss = tf.reduce_mean(
tf.square(
tf.subtract(
tf.multiply(
COLOR_MAX,
nextBatch
),
tf.multiply(
COLOR_MAX,
decompressed
)
)
)
)
# Optimize error for next batch
optimizer = tf.contrib.optimizer_v2.AdamOptimizer(0.001)
minimizer = optimizer.minimize(loss)
tf.global_variables_initializer().run()
saver=tf.train.Saver(max_to_keep=2)
def voxelSaver(voxels, iteration, loss_out):
if (iteration_index == 0):
bat = nextBatch.eval()
reco = decompressed.eval()
bat = (bat *255).astype(np.uint8)
reco = (reco *255).astype(np.uint8)
# Save the original
np.save(
"data/%i_original_loss_%s_reluout.npy" % (
iteration, loss_out
), bat)
print("Original min %i, max %s, median %d " % (
np.min(bat), np.max(bat), np.median(bat)
))
# save the reconstruction
np.save(
"data/%i_reconstructed_loss_%s_reluout.npy" % (
iteration, loss_out
), reco)
print("Reconstructed min %i, max %s, median %d " % (
np.min(reco), np.max(reco), np.median(reco)
))
else :
reco = decompressed.eval()
reco = (reco *255).astype(np.uint8)
# save the reconstruction
np.save(
"data/%i_reconstructed_loss_%s_reluout.npy" % (
iteration, loss_out
), reco)
print("Reconstructed min %i, max %s, median %d " % (
np.min(reco), np.max(reco), np.median(reco)
))
print("=====Voxels saved!!======")
def process(voxels):
global iteration_index
iterator.initializer.run(feed_dict = {feed : voxels})
minimizer.run(feed_dict={feed : voxels})
print("Iteration %i" % iteration_index)
if (iteration_index <= 7000 and (iteration_index % 58 == 0)):
loss_out = loss.eval()
print("=====loss at %i: %s=======" % (iteration_index, loss_out))
a = [n.name for n in tf.get_default_graph().as_graph_def().node]
print(len(a))
if (iteration_index == 0):
voxelSaver(voxels, iteration_index, loss_out)
if (loss_out <= 500 and iteration_index % 116 == 0):
voxelSaver(voxels, iteration_index, loss_out)
saver.save(
sess, SAVEPATH)
print("Graph saved successfully.")
elif (7000 < iteration_index <= 10000 and iteration_index % 116 == 0):
loss_out = loss.eval()
print("=====loss at %i: %s=======" % (iteration_index, loss_out))
if (loss_out <= 350):
voxelSaver(voxels, iteration_index, loss_out)
saver.save(
sess, SAVEPATH)
print("Graph saved successfully.")
elif (10000 < iteration_index <= 50000 and iteration_index % 116 == 0):
loss_out = loss.eval()
print("=====loss at %i: %s=======" % (iteration_index, loss_out))
if (loss_out <= 200):
voxelSaver(voxels, iteration_index, loss_out)
saver.save(
sess, SAVEPATH)
print("Graph saved successfully.")
elif (50000 < iteration_index and iteration_index % 1160 == 0):
loss_out = loss.eval()
print("=====loss at %i: %s=======" % (iteration_index, loss_out))
if (loss_out <= 100):
voxelSaver(voxels, iteration_index, loss_out)
saver.save(
sess, SAVEPATH)
print("Graph saved successfully.")
else:
pass
iteration_index += 1