-
Notifications
You must be signed in to change notification settings - Fork 1
/
watch.py
375 lines (357 loc) · 18.4 KB
/
watch.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
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
import cv2
import matplotlib.patches as mpatches
import matplotlib.pyplot as plt
import numpy as np
import os
import sys
from worlds.base_world import World as BaseWorld
import core.tools as tools
import becca_tools_control_panel.control_panel as cp
import worlds.world_tools as wtools
class World(BaseWorld):
""" The watch world provides a sequence of video frames to the BECCA agent
There are no actions that the agent can take that affect the world.
This world uses the OpenCV library. Installation instructions are here:
http://docs.opencv.org/doc/tutorials/introduction/linux_install/linux_install.html#linux-installation
"""
# This package assumes that it is located directly under the BECCA package
def __init__(self, lifespan=None):
super(World, self).__init__()
if lifespan is not None:
self.LIFESPAN = lifespan
# Flag indicates whether the world is in testing mode
#self.short_test = False
self.TEST = False
self.VISUALIZE_PERIOD = 10 ** 1
# Flag determines whether to plot all the features during display
self.print_all_features = True
self.fov_horz_span = 16
self.fov_vert_span = 12
self.name = 'watch_world_12x16'
print "Entering", self.name
# Generate a list of the filenames to be used
self.video_filenames = []
extensions = ['.mpg', '.mp4', '.flv', '.avi']
if self.TEST:
test_filename = 'test_long.avi'
truth_filename = 'truth_long.txt'
self.video_filenames = []
self.video_filenames.append(os.path.join(
'becca_world_watch', 'test', test_filename))
self.ground_truth_filename = os.path.join('becca_world_watch',
'test', truth_filename)
else:
self.data_dir_name = os.path.join('becca_world_watch', 'data')
self.video_filenames = tools.get_files_with_suffix(
self.data_dir_name, extensions)
self.video_file_count = len(self.video_filenames)
print self.video_file_count, 'video files loaded.'
# Initialize the video data to be viewed
self.initialize_video_file()
self.num_sensors = 2 * self.fov_horz_span * self.fov_vert_span
self.num_actions = 0
self.initialize_control_panel()
self.frame_counter = 10000
self.frames_per_step = 3
if self.TEST:
self.surprise_log_filename = os.path.join('becca_world_watch',
'log', 'surprise.txt')
self.surprise_log = open(self.surprise_log_filename, 'w')
def initialize_video_file(self):
""" Queue up one of the video files and get it ready for processing """
filename = self.video_filenames[
np.random.randint(0, self.video_file_count)]
print 'Loading', filename
self.video_reader = cv2.VideoCapture(filename)
self.clip_frame = 0
def step(self, action):
""" Advance the video one time step and read and process the frame """
for _ in range(self.frames_per_step):
((success, image)) = self.video_reader.read()
# Check whether the end of the clip has been reached
if not success:
if self.TEST:
# Terminate the test
self.video_reader.release()
self.surprise_log.close()
print 'End of test reached'
tools.report_roc(self.ground_truth_filename,
self.surprise_log_filename, self.name)
sys.exit()
else:
self.initialize_video_file()
((success, image)) = self.video_reader.read()
self.timestep += 1
self.clip_frame += self.frames_per_step
image = image.astype('float') / 256.
# Convert the color image to grayscale
self.intensity_image = np.sum(image, axis=2) / 3.
# Convert the grayscale to center-surround contrast pixels
center_surround_pixels = wtools.center_surround(
self.intensity_image, self.fov_horz_span, self.fov_vert_span)
unsplit_sensors = center_surround_pixels.ravel()
self.sensors = np.concatenate((np.maximum(unsplit_sensors, 0),
np.abs(np.minimum(unsplit_sensors, 0))))
reward = 0
return self.sensors, reward
def set_agent_parameters(self, agent):
""" Manually set some agent parameters, where required """
agent.VISUALIZE_PERIOD = self.VISUALIZE_PERIOD
if self.TEST:
# Prevent the agent from adapting during testing
agent.BACKUP_PERIOD = 10 ** 9
for block in agent.blocks:
block.ziptie.COACTIVITY_UPDATE_RATE = 0.
block.ziptie.JOINING_THRESHOLD = 2.
block.ziptie.AGGLOMERATION_ENERGY_RATE = 0.
block.ziptie.NUCLEATION_ENERGY_RATE = 0.
for cog in block.cogs:
cog.ziptie.COACTIVITY_UPDATE_RATE = 0.
cog.ziptie.JOINING_THRESHOLD = 2.
cog.ziptie.AGGLOMERATION_ENERGY_RATE = 0.
cog.ziptie.NUCLEATION_ENERGY_RATE = 0.
cog.daisychain.CHAIN_UPDATE_RATE = 0.
else:
pass
return
def initialize_control_panel(self):
""" Prepare the user display of the world's internal state """
self.fig = cp.figure()
self.ax_original_image = cp.subfigure(self.fig,
left=0., bottom=0.4, width=0.45, height=0.6)
self.ax_sensed_image = cp.subfigure(self.fig,
left=0., bottom=0., width=0.3, height=0.4)
self.ax_interpreted_image = cp.subfigure(self.fig,
left=0.3, bottom=0., width=0.3, height=0.4)
self.ax_status = cp.subfigure(self.fig,
left=0.45, bottom=0.4, width=0.15, height=0.6)
# Initialize original image
plt.gray()
self.original_image = self.ax_original_image.imshow(
np.zeros((self.fov_vert_span, self.fov_horz_span)),
vmin=0., vmax=1., interpolation='nearest',
animated=True)
self.ax_original_image.text(-.01, -.01, 'Original image',
size=10, color=tools.OXIDE,
ha='left', va='center')
self.ax_original_image.get_xaxis().set_visible(False)
self.ax_original_image.get_yaxis().set_visible(False)
# Initialize sensed image
plt.gray()
self.sensed_image = self.ax_sensed_image.imshow(
np.zeros((self.fov_vert_span, self.fov_horz_span)),
vmin=0., vmax=1., interpolation='nearest',
animated=True)
self.ax_sensed_image.text(-.01, -.01, 'Sensed image',
size=10, color=tools.OXIDE,
ha='left', va='center')
self.ax_sensed_image.get_xaxis().set_visible(False)
self.ax_sensed_image.get_yaxis().set_visible(False)
# Initialize interpreted image
plt.gray()
self.interpreted_image = self.ax_interpreted_image.imshow(
np.zeros((self.fov_vert_span, self.fov_horz_span)),
vmin=0., vmax=1., interpolation='nearest',
animated=True)
self.ax_interpreted_image.text(-.01, -.01, 'Interpreted image',
size=10, color=tools.OXIDE,
ha='left', va='center')
self.ax_interpreted_image.get_xaxis().set_visible(False)
self.ax_interpreted_image.get_yaxis().set_visible(False)
# Initialize status window
self.ax_status.axis((0., 1., 0., 1.))
self.ax_status.get_xaxis().set_visible(False)
self.ax_status.get_yaxis().set_visible(False)
self.clip_time_status = self.ax_status.text(-0.05, 0.13,
'Clip time:',
color=tools.COPPER_SHADOW, size=10, ha='left', va='center')
self.wake_time_status = self.ax_status.text(-0.05, 0.08,
'Wake time:',
color=tools.COPPER_SHADOW, size=10, ha='left', va='center')
self.life_time_status = self.ax_status.text(-0.05, 0.03,
'Life time:',
color=tools.COPPER_SHADOW, size=10, ha='left', va='center')
self.surprise_status = self.ax_status.text(-0.05, 0.4,
'Novelty: ',
color=tools.COPPER_SHADOW, size=10, ha='left', va='center')
# Initialize surprise plot
self.surprise_ax_left = 0.6
self.surprise_ax_bottom = 0.56
self.surprise_ax_width = 0.4
self.surprise_ax_height = 0.44
self.ax_surprise = cp.subfigure(self.fig, left=self.surprise_ax_left,
bottom=self.surprise_ax_bottom,
width=self.surprise_ax_width,
height=self.surprise_ax_height)
self.ax_surprise.axis((0., 1., 0., 1.))
self.ax_surprise.get_xaxis().set_visible(False)
self.ax_surprise.get_yaxis().set_visible(False)
self.block_ax_vert_border = 0.02 * self.surprise_ax_height
self.block_ax_horz_border = 0.04 * self.surprise_ax_width
self.surprise_block_ax = []
# Initialize features plot
self.feature_ax_left = 0.6
self.feature_ax_bottom = 0.12
self.feature_ax_width = 0.4
self.feature_ax_height = 0.44
self.ax_features = cp.subfigure(self.fig, left=self.feature_ax_left,
bottom=self.feature_ax_bottom,
width=self.feature_ax_width,
height=self.feature_ax_height)
self.ax_features.axis((0., 1., 0., 1.))
self.ax_features.get_xaxis().set_visible(False)
self.ax_features.get_yaxis().set_visible(False)
self.feature_ax_vert_border = 0.028 * self.feature_ax_height
self.feature_ax_horz_border = 0.005 * self.feature_ax_width
self.block_ax = []
self.fig.show()
def visualize(self, agent):
""" Update the display to the user of the world's internal state """
if self.TEST:
# Save the surprise value
surprise_val = agent.surprise_history[-1]
time_in_seconds = str(float(self.clip_frame) / 30.)
file_line = ' '.join([str(surprise_val), str(time_in_seconds)])
self.surprise_log.write(file_line)
self.surprise_log.write('\n')
if (self.timestep % self.VISUALIZE_PERIOD != 0):
return
print self.timestep, 'steps'
(projections, feature_activities) = agent.get_projections()
# Make a copy of projections for finding the interpretation
interpretation_by_feature = list(projections)
interpretation = np.zeros((self.num_sensors, 1))
for block_index in range(len(interpretation_by_feature)):
for feature_index in range(len(interpretation_by_feature
[block_index])):
this_feature_interpretation = (
interpretation_by_feature[block_index]
[feature_index][:self.num_sensors,-1][:,np.newaxis])
interpretation = np.maximum(interpretation,
this_feature_interpretation *
feature_activities[block_index][feature_index])
self.original_image.set_data(self.intensity_image)
sensed_image_array = wtools.visualize_pixel_array_feature(
self.sensors[:,np.newaxis], fov_horz_span=self.fov_horz_span,
fov_vert_span=self.fov_vert_span, array_only=True)
self.sensed_image.set_data(sensed_image_array[0])
interpreted_image_array = wtools.visualize_pixel_array_feature(
interpretation[:self.num_sensors],
fov_horz_span=self.fov_horz_span,
fov_vert_span=self.fov_vert_span, array_only=True)
self.interpreted_image.set_data(interpreted_image_array[0])
# Update status window
self.clip_time_status.set_text(''.join((
'Clip time: ', '%0.2f' % (self.clip_frame/(30.*60.)), ' min')))
self.wake_time_status.set_text(''.join((
'Wake time: ', '%0.2f' % (self.timestep * self.frames_per_step
/ (30.*60.)), ' min')))
self.life_time_status.set_text(''.join((
'Life time: ', '%0.2f' % (agent.timestep * self.frames_per_step
/ (30.*60.)), ' min')))
self.surprise_status.set_text(''.join((
'Novelty: ', '%0.2f' % agent.surprise_history[-1])))
# Update surprise visualization window
# Clear all axes
for axes in self.surprise_block_ax:
self.fig.delaxes(axes)
self.surprise_block_ax = []
# Display each block's features and bundle activities.
# The top block has no bundles.
num_blocks = len(agent.blocks)
for block_index in range(num_blocks):
block = agent.blocks[block_index]
block_surprise = block.surprise
num_cogs_in_block = len(block.cogs)
surprise_array = np.reshape(block_surprise,
(num_cogs_in_block,
block.max_bundles_per_cog)).T
block_left = self.surprise_ax_left + self.block_ax_horz_border
block_height = ((self.surprise_ax_height -
self.block_ax_vert_border -
self.feature_ax_vert_border * 2) /
float(num_blocks) -
self.block_ax_vert_border)
block_bottom = (self.surprise_ax_bottom +
self.feature_ax_vert_border +
self.block_ax_vert_border +
(block_height + self.block_ax_vert_border) *
block_index)
block_width = (self.surprise_ax_width -
2 * self.block_ax_horz_border)
last_block_top = block_bottom + block_height
rect = (block_left, block_bottom, block_width, block_height)
ax = self.fig.add_axes(rect, frame_on=False)
ax.get_xaxis().set_visible(False)
ax.get_yaxis().set_visible(False)
plt.gray()
im = ax.imshow(surprise_array, aspect='auto',
interpolation='nearest', vmin=0., vmax=1.,
cmap='copper')
if block_index == 0:
ax.text(num_cogs_in_block * 0.85,
block.max_bundles_per_cog * 0.8,
'Novelty', color=tools.OXIDE,
size=10, ha='left', va='bottom')
self.surprise_block_ax.append(ax)
# Update feature visualization window
# Clear all axes
for axes in self.block_ax:
self.fig.delaxes(axes)
self.block_ax = []
# Display each block's features and bundle activities.
# The top block has no bundles.
num_blocks = len(agent.blocks)
for block_index in range(num_blocks):
block = agent.blocks[block_index]
cable_activities = block.cable_activities
num_cogs_in_block = len(block.cogs)
activity_array = np.reshape(cable_activities,
(num_cogs_in_block,
block.max_bundles_per_cog)).T
block_left = self.feature_ax_left + self.block_ax_horz_border
block_height = ((self.feature_ax_height -
self.block_ax_vert_border -
self.feature_ax_vert_border * 2) /
float(num_blocks) -
self.block_ax_vert_border)
block_bottom = (self.feature_ax_bottom +
self.feature_ax_vert_border +
self.block_ax_vert_border +
(block_height + self.block_ax_vert_border) *
block_index)
block_width = self.feature_ax_width - \
2 * self.block_ax_horz_border
last_block_top = block_bottom + block_height
rect = (block_left, block_bottom, block_width, block_height)
ax = self.fig.add_axes(rect, frame_on=False)
ax.get_xaxis().set_visible(False)
ax.get_yaxis().set_visible(False)
plt.gray()
im = ax.imshow(activity_array, aspect='auto',
interpolation='nearest', vmin=0., vmax=1.,
cmap='copper')
if block_index == 0:
ax.text(num_cogs_in_block * 0.85,
block.max_bundles_per_cog * 0.8,
'Activities', color=tools.OXIDE,
size=10, ha='left', va='bottom')
self.block_ax.append(ax)
if self.print_all_features:
log_directory = os.path.join('becca_world_watch', 'log')
wtools.print_pixel_array_features(projections, self.num_sensors,
self.num_actions,
self.fov_horz_span,
self.fov_vert_span,
directory=log_directory,
world_name=self.name)
self.fig.canvas.draw()
plt.draw()
# Save the control panel image
filename = self.name + '_' + str(self.frame_counter) + '.png'
full_filename = os.path.join('becca_world_watch', 'frames', filename)
self.frame_counter += 1
plt.figure(self.fig.number)
#plt.savefig(full_filename, format='png', dpi=80) # for 720
plt.savefig(full_filename, format='png', dpi=120) # for 1080
return