-
Notifications
You must be signed in to change notification settings - Fork 156
/
agent.py
executable file
·260 lines (229 loc) · 8.98 KB
/
agent.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
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
import tensorflow as tf
import tensorflow.contrib.layers as ly
from util import lrelu
import cv2
import math
from pdf_sample_layer import pdf_sample
from util import enrich_image_input
from util import STATE_DROPOUT_BEGIN, STATE_REWARD_DIM, STATE_STEP_DIM, STATE_STOPPED_DIM
def feature_extractor(net, output_dim, cfg):
net = net - 0.5
min_feature_map_size = 4
assert output_dim % (
min_feature_map_size**2) == 0, 'output dim=%d' % output_dim
size = int(net.get_shape()[2])
print('Agent CNN:')
channels = cfg.base_channels
print(' ', str(net.get_shape()))
size /= 2
net = ly.conv2d(
net, num_outputs=channels, kernel_size=4, stride=2, activation_fn=lrelu)
print(' ', str(net.get_shape()))
while size > min_feature_map_size:
if size == min_feature_map_size * 2:
channels = output_dim / (min_feature_map_size**2)
else:
channels *= 2
assert size % 2 == 0
size /= 2
net = ly.conv2d(
net, num_outputs=channels, kernel_size=4, stride=2, activation_fn=lrelu)
print(' ', str(net.get_shape()))
print('before fc: ', net.get_shape()[1])
net = tf.reshape(net, [-1, output_dim])
net = tf.nn.dropout(net, cfg.dropout_keep_prob)
return net
# Output: float \in [0, 1]
def agent_generator(inp, is_train, progress, cfg, high_res=None, alex_in=None):
net, z, states = inp
filters = cfg.filters
filters = [x(net, cfg) for x in filters]
selection_noise = z[:, 0:1]
filtered_images = []
filter_debug_info = []
high_res_outputs = []
if cfg.shared_feature_extractor:
filter_features = feature_extractor(
net=enrich_image_input(cfg, net, states),
output_dim=cfg.feature_extractor_dims,
cfg=cfg)
# filter_features = ly.dropout(filter_features)
for j, filter in enumerate(filters):
with tf.variable_scope('filter_%d' % j):
print(' creating filter:', j, 'name:', str(filter.__class__), 'abbr.',
filter.get_short_name())
if not cfg.shared_feature_extractor:
filter_features = \
feature_extractor(net=enrich_image_input(cfg, net),
output_dim=cfg.feature_extractor_dims, cfg=cfg)
print(' filter_features:', filter_features.shape)
filtered_image_batch, high_res_output, per_filter_debug_info = filter.apply(
net, filter_features, high_res=high_res)
high_res_outputs.append(high_res_output)
filtered_images.append(filtered_image_batch)
filter_debug_info.append(per_filter_debug_info)
print(' output:', filtered_image_batch.shape)
# [batch_size, #filters, H, W, C]
for img in filtered_images:
print('img', img.shape)
filtered_images = tf.stack(values=filtered_images, axis=1)
print(' filtered_images:', filtered_images.shape)
with tf.variable_scope('action_selection'):
selector_features = feature_extractor(
net=enrich_image_input(cfg, net, states),
output_dim=cfg.feature_extractor_dims,
cfg=cfg)
print(' selector features:', selector_features.shape)
selector_features = ly.fully_connected(
selector_features,
num_outputs=cfg.fc1_size,
scope='selector_fc1',
activation_fn=lrelu)
# selector_features = ly.dropout(selector_features)
pdf = ly.fully_connected(
selector_features,
num_outputs=len(filters),
activation_fn=None,
scope='selector_fc2')
pdf = tf.nn.softmax(pdf) + 1e-37
print(' pdf_filter', pdf[:, 1:].shape)
# print(' pdf_mask', states[:, STATE_DROPOUT_BEGIN:].shape)
pdf = pdf * (1 - cfg.exploration) + cfg.exploration * 1.0 / len(filters)
# pdf = tf.to_float(is_train) * tf.concat([pdf[:, :1], pdf[:, 1:] * states[:, STATE_DROPOUT_BEGIN:]], axis=1) \
# + (1.0 - tf.to_float(is_train)) * pdf
pdf = pdf / (tf.reduce_sum(pdf, axis=1, keep_dims=True) + 1e-30)
entropy = -pdf * tf.log(pdf)
entropy = tf.reduce_sum(entropy, axis=1)[:, None]
print(' pdf:', pdf.shape)
print(' entropy:', entropy.shape)
print(' selection_noise:', selection_noise.shape)
random_filter_id = pdf_sample(pdf, selection_noise)
max_filter_id = tf.cast(tf.argmax(pdf, axis=1), tf.int32)
selected_filter_id = is_train * random_filter_id + (
1 - is_train) * max_filter_id
print(' selected_filter_id:', selected_filter_id.shape)
filter_one_hot = tf.one_hot(
selected_filter_id, depth=len(filters), dtype=tf.float32)
print(' filter one_hot', filter_one_hot.shape)
surrogate = tf.reduce_sum(
filter_one_hot * tf.log(pdf + 1e-10), axis=1, keep_dims=True)
net = tf.reduce_sum(
filtered_images * filter_one_hot[:, :, None, None, None], axis=1)
if high_res is not None:
high_res_outputs = tf.stack(values=high_res_outputs, axis=1)
high_res_output = tf.reduce_sum(
high_res_outputs * filter_one_hot[:, :, None, None, None], axis=1)
# only the first image will get debug_info
debug_info = {
'state': states,
'selected_filter_id': selected_filter_id[0],
'filter_debug_info': filter_debug_info,
'pdf': pdf[0]
}
# Combined: Three in one 64x64 ?
# otherwise returns pdf, detail, mask
def debugger(debug_info, combined=True):
size = 8
img = None
images = [None for i in range(3)]
for i, filter in enumerate(filters):
selected = i == debug_info['selected_filter_id']
if selected:
img = filter.visualize_mask(debug_info['filter_debug_info'][i],
(64, 64)) * 0.8
assert img is not None
if not combined:
# Mask
images[2] = img.copy()
# reset img
img = img * 0 + 0.5
c = 0
for i, filter in enumerate(filters):
pdf = debug_info['pdf'][i]
if pdf < 1e-10:
continue
else:
c += 1
selected = i == debug_info['selected_filter_id']
if selected:
filter.visualize_filter(debug_info['filter_debug_info'][i], img)
if not combined:
# detail
images[1] = img.copy()
# reset img
img = img * 0 + 0.5
c = 0
for i, filter in enumerate(filters):
per_col = 4
x = c // per_col * 30
y = size * (c % per_col + 1)
pdf = debug_info['pdf'][i]
if pdf < 1e-10:
continue
else:
c += 1
cv2.putText(img,
filter.get_short_name(), (x + 6, y + 4),
cv2.FONT_HERSHEY_SIMPLEX, 0.233, (255, 255, 255))
selected = i == debug_info['selected_filter_id']
color = 1.0 if selected else 0.3
width = int(pdf * 20)
height = 0.35
corners = [(x + 16, int(y + (1 - height) * size // 2)),
(x + 16 + width, int(y + (1 + height) * size // 2))]
cv2.rectangle(img, (corners[0][0] - 1, corners[0][1] - 1),
(corners[1][0] + 1, corners[1][1] + 1), (1, 1,
1), cv2.FILLED)
cv2.rectangle(img, corners[0], corners[1], (color, 0.3, 0.3), cv2.FILLED)
if not combined:
# pdf
images[0] = img.copy()
if combined:
return img
else:
return images
debugger.width = int(net.shape[1])
print(' surrogate: ', surrogate.shape)
# Calculate new states
new_states = [None for _ in range(STATE_DROPOUT_BEGIN + 1)]
is_last_step = tf.cast(
tf.abs(states[:, STATE_STEP_DIM:STATE_STEP_DIM + 1] + 1 - cfg.test_steps)
< 1e-4,
dtype=tf.float32)
submitted = is_last_step
new_states[STATE_REWARD_DIM] = submitted
new_states[STATE_STOPPED_DIM] = submitted
# Increment the step
new_states[STATE_STEP_DIM] = (states[:, STATE_STEP_DIM] + 1)[:, None]
# Update filter usage
filter_usage = states[:, STATE_STEP_DIM + 1:]
print('usage v.s. onehot', filter_usage.shape, filter_one_hot.shape)
assert len(filter_usage.shape) == len(filter_one_hot.shape)
regular_filter_start = 0
# Penalize submission action that is not the final action.
early_stop_penalty = (1 - is_last_step) * submitted * cfg.early_stop_penalty
usage_penalty = tf.reduce_sum(
filter_usage * filter_one_hot[:, regular_filter_start:],
axis=1,
keep_dims=True)
new_filter_usage = tf.maximum(filter_usage,
filter_one_hot[:, regular_filter_start:])
new_states[STATE_STEP_DIM + 1] = new_filter_usage
print(submitted.shape, new_states[STATE_STEP_DIM].shape)
new_states = tf.concat(new_states, axis=1)
print('new_states:', new_states.shape)
if cfg.clamp:
net = tf.clip_by_value(net, 0.0, 5.0)
entropy_penalty = (1.0 - progress) * cfg.exploration_penalty * (
-entropy + math.log(len(filters)))
# Will be substracted from award
penalty = tf.reduce_mean(
tf.maximum(net - 1, 0)**2, axis=(1, 2, 3)
)[:,
None] + entropy_penalty + usage_penalty * cfg.filter_usage_penalty + early_stop_penalty
print('states, new_states:', states.shape, new_states.shape)
print('penalty:', penalty.shape)
if high_res is None:
return (net, new_states, surrogate, penalty), debug_info, debugger
else:
return (net, new_states, high_res_output), debug_info, debugger