-
Notifications
You must be signed in to change notification settings - Fork 148
/
utils.py
644 lines (534 loc) · 21.9 KB
/
utils.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
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
from __future__ import (absolute_import, division,
print_function, unicode_literals)
import math
import re
# TODO: Bootstrap future module to enable Python 2 support of install which depends on this file to do below
# from future.builtins import (ascii, bytes, chr, dict, filter, hex, input,
# int, map, next, oct, open, pow, range, round,
# str, super, zip)
import ctypes
import platform
import shutil
import glob
import inspect
import os
import sys
import threading
import time
import traceback
from os.path import exists, expanduser, basename
from typing import Tuple
import numpy as np
import h5py
import requests
from box import Box
import config as c
import logs
from util.anonymize import anonymize_user_home
from util.download import download
from util.ensure_sim import ensure_sim
from util.run_command import run_command
log = logs.get_log(__name__)
def normalize(a):
amax = a.max()
amin = a.min()
arange = amax - amin
a = (a - amin) / arange
return a
def preprocess_image(image):
start = time.time()
image = (image.astype(np.float32, copy=False)
** 0.45 # gamma correct
* 255.)
image = np.clip(image, a_min=0, a_max=255)\
.astype('uint8', copy=False)
end = time.time()
log.debug('preprocess_capture_image took %rms', (end - start) * 1000.)
return image
def preprocess_depth(depth):
depth = depth.astype('float64', copy=False)
# x = list(range(depth.size))
# y = depth.flatten()
# plt.scatter(x, y)
# plt.show()
depth = depth ** -(1 / 3.)
depth = normalize(depth)
return depth
def depth_heatmap(depth):
red = depth
green = 1.0 - np.abs(0.5 - depth) * 2.
blue = 1. - depth
ret = np.array([red, green, blue])
ret = np.transpose(ret, (1, 2, 0))
ret = (ret * 255).astype('uint8', copy=False)
return ret
def obj2dict(obj, exclude=None):
"""
Converts object properties to a dict.
This acts as a single level copy, i.e. it's NOT recursive.
@:param obj - The Object to convert
@:param exclude - A list of property names to omit from the returned object
"""
ret = {}
exclude = exclude or []
for name in dir(obj):
if not name.startswith('__') and name not in exclude:
value = getattr(obj, name)
if not callable(value):
ret[name] = value
return ret
def save_hdf5(out, filename, background=True):
assert_disk_space(os.path.dirname(filename))
if 'DEEPDRIVE_NO_THREAD_SAVE' in os.environ or not background:
return save_hdf5_task(out, filename)
else:
thread = threading.Thread(target=save_hdf5_task, args=(out, filename))
thread.start()
return thread
def save_hdf5_task(out, filename):
os.makedirs(os.path.dirname(filename), exist_ok=True)
log.debug('Saving to %s', filename)
opts = dict(compression='lzf', fletcher32=True)
with h5py.File(filename, 'w') as f:
for i, frame in enumerate(out):
frame_grp = f.create_group('frame_%s' % str(i).zfill(10))
add_collision_to_hdf5(frame, frame_grp)
add_return_to_hdf5(frame, frame_grp)
add_world_to_hdf5(frame, frame_grp)
add_cams_to_hdf5(frame, frame_grp, opts)
del frame['cameras']
for k, v in frame.items():
frame_grp.attrs[k] = v
log.info('Saved to %s', filename)
def add_world_to_hdf5(frame, frame_grp):
parent_key = 'world'
world = frame[parent_key]
world_grp = frame_grp.create_group(parent_key)
for k, v in world.items():
if k == 'vehicle_positions':
v = np.array(v)
world_grp.attrs[k] = v
del frame[parent_key]
def add_cams_to_hdf5(frame, frame_grp, opts):
for j, camera in enumerate(frame['cameras']):
camera_grp = frame_grp.create_group('camera_%s' % str(j).zfill(5))
camera_grp.create_dataset('image', data=camera['image'], **opts)
camera_grp.create_dataset('depth', data=camera['depth'], **opts)
del camera['image_data']
del camera['depth_data']
del camera['image']
if 'image_raw' in camera:
del camera['image_raw']
del camera['depth']
for k, v in camera.items():
# TODO: Move this to a 'props' dataset as attrs can only be 64kB
camera_grp.attrs[k] = v
def add_return_to_hdf5(frame, frame_grp):
from sim.return_aggregator import EpisodeReturn
episode_return = frame['episode_return']
return_grp = frame_grp.create_group('episode_return')
defaults = obj2dict(EpisodeReturn)
prop_names = defaults.keys()
for k in prop_names:
if 'sampler' not in k.lower():
return_grp.attrs[k] = episode_return.get(k, defaults[k])
del frame['episode_return']
def add_collision_to_hdf5(frame, frame_grp):
from box import Box
clsn_grp = frame_grp.create_group('last_collision')
clsn = Box(frame['last_collision'], box_it_up=True)
clsn_grp.attrs['collidee_velocity'] = tuple(clsn.collidee_velocity)
collidee_location = getattr(clsn, 'collidee_location', None)
clsn_grp.attrs['collidee_location'] = \
collidee_location if (clsn.time_utc and collidee_location) else ''
clsn_grp.attrs['collision_normal'] = tuple(clsn.collision_normal)
clsn_grp.attrs['time_since_last_collision'] = clsn.time_since_last_collision
clsn_grp.attrs['time_stamp'] = clsn.time_stamp
clsn_grp.attrs['time_utc'] = clsn.time_utc
del frame['last_collision']
def read_hdf5(filename, save_png_dir=None, overfit=False, save_prefix=''):
ret = []
with h5py.File(filename, 'r') as file:
for i, frame_name in enumerate(file):
out_frame = read_frame(file, frame_name, i, save_png_dir,
save_prefix)
if out_frame is None:
log.error('Could not read frame, skipping')
else:
ret.append(out_frame)
if overfit:
log.info('overfitting to %r, image# %d', filename, i)
if i == 1:
break
return ret
def read_frame(file, frame_name, frame_index, save_png_dir, save_prefix=''):
try:
frame = file[frame_name]
out_frame = dict(frame.attrs)
out_cameras = []
for dataset_name in frame:
if dataset_name.startswith('camera_'):
read_camera(dataset_name, frame, frame_index,
out_cameras, save_png_dir, save_prefix)
elif dataset_name == 'last_collision':
out_frame['last_collision'] = dict(frame[dataset_name].attrs)
out_frame['cameras'] = out_cameras
except Exception as e:
traceback.print_stack()
log.error('Exception reading frame %s', str(e))
out_frame = None
return out_frame
def read_camera(dataset_name, frame, frame_index, out_cameras, save_png_dir,
save_prefix=''):
camera = frame[dataset_name]
out_camera = dict(camera.attrs)
out_camera['image'] = camera['image'][()]
out_camera['depth'] = camera['depth'][()]
out_cameras.append(out_camera)
if save_png_dir is not None:
if not os.path.exists(save_png_dir):
os.makedirs(save_png_dir)
save_camera(out_camera['image'], out_camera['depth'],
save_dir=save_png_dir, name=save_prefix + str(frame_index)
.zfill(c.HDF5_FRAME_ZFILL))
def save_camera(image, depth, save_dir, name):
from scipy.misc import imsave
im_path = os.path.join(save_dir, 'i_' + name + '.png')
dp_path = os.path.join(save_dir, 'z_' + name + '.png')
imsave(im_path, image)
imsave(dp_path, depth)
log.debug('saved image and depth to %s and %s', im_path, dp_path)
def show_camera(image, depth):
from scipy.misc import toimage
toimage(image).show()
toimage(depth).show()
input('Enter any key to continue')
def hdf5_to_mp4(fps=c.DEFAULT_FPS, png_dir=None, combine_all=False, sess_dir=None):
if png_dir is None:
png_dir = save_hdf5_recordings_to_png(combine_all, sess_dir)
try:
file_path = pngs_to_mp4(combine_all, fps, png_dir)
finally:
guarded_rmtree(png_dir)
return file_path
def pngs_to_mp4(combine_all, fps, png_dir):
# TODO: Add FPS, frame number, run id, date str,
# g-forces, episode #, hdf5 #, etc... to this
# and rendered views for human interprettability
log.info('Saved png\'s to ' + png_dir)
file_path = None
import distutils.spawn
ffmpeg_path = distutils.spawn.find_executable('ffmpeg')
if ffmpeg_path is None:
log.error('Could not find ffmpeg. Skipping hdf5=>mp4 conversion')
else:
zfill_total = c.HDF5_DIR_ZFILL + c.HDF5_FRAME_ZFILL
pix_fmt = 'yuv420p' # The pix_fmt does not define resolution (i.e. this is totally different than 480p)
title = 'deepdrive'
file_dir = c.RESULTS_DIR
if not combine_all:
title += '_' + c.DATE_STR
file_path = os.path.join(file_dir, '%s.mp4' % title)
ffmpeg_cmd = ('ffmpeg'
' -y '
' -r {fps}'
' -f image2'
' -i {temp_png_dir}/i_hdf5_%0{zfill_total}d.png'
' -vcodec libx264'
' -crf 25'
' -pix_fmt {pix_fmt}'
' -vf "pad=ceil(iw/2)*2:ceil(ih/2)*2"'
' {file_path}'.format(fps=fps, pix_fmt=pix_fmt,
zfill_total=zfill_total,
file_path=file_path,
temp_png_dir=png_dir))
log.info('PNG=>MP4: ' + ffmpeg_cmd)
ffmpeg_result = os.system(ffmpeg_cmd)
if ffmpeg_result == 0:
log.info('Wrote mp4 to: ' + anonymize_user_home(file_path))
else:
file_path = None
return file_path
def upload_to_gist(name: str, file_paths: list, public: bool):
files = ' '.join('"%s"' % f for f in file_paths)
gist_env = os.environ.copy()
gist_env['YOU_GET_MY_JIST'] = requests.get(c.YOU_GET_MY_JIST_URL).text.strip()
if os.path.dirname(sys.executable) not in os.environ['PATH']:
gist_env['PATH'] = os.path.dirname(sys.executable) + ':' + gist_env['PATH']
opts = '--public' if public else ''
cmd = 'gist {opts} create {gist_name} {files}'
cmd = cmd.format(gist_name=name, files=files, opts=opts)
output, ret_code = run_command(cmd, env=gist_env, verbose=True)
if ret_code != 0:
log.warn('Could not upload gist. \n%s' % (output,))
url = output if ret_code == 0 else None
return url
def in_home(name):
p = os.path
return p.exists(p.join(p.expanduser('~'), name))
def upload_to_youtube(file_path):
youtube_creds_name = '.youtube-upload-credentials.json'
client_secrets_name = '.client_secrets.json'
youtube_creds_exists = in_home(youtube_creds_name)
client_secrets_exists = in_home(client_secrets_name)
if not youtube_creds_exists or not client_secrets_exists:
log.error('Need %s and %s in your home directory to upload to YouTube.',
youtube_creds_name, client_secrets_name)
return False
# python_path = os.environ['PYTHONPATH']
# youtube_upload_dir = os.path.join(c.ROOT_DIR, 'vendor', 'youtube_upload')
# os.environ['PYTHONPATH'] = '%s:%s' % (youtube_upload_dir, python_path)
import youtube_upload.main
_, options, _ = youtube_upload.main.get_options([])
options._update_careful(dict(
default_box=True,
title=basename(file_path), privacy='unlisted', client_secrets='',
credentials_file='', auth_browser=None,
description='Deepdrive results for %s' % c.MAIN_ARGS))
youtube = youtube_upload.main.get_youtube_handler(options)
video_id = youtube_upload.main.upload_youtube_video(youtube, options,
file_path, 1, 0)
# TODO: Put link to s3 artifacts in description [hdf5, csv, diff,
# eventually ue-recording]
# cmd = '%s %s --title=test --privacy=unlisted %s' % (
# sys.executable,
# os.path.join(youtube_upload_dir, 'bin', 'youtube_upload'),
# file_path
# )
# os.environ['PYTHONPATH'] = python_path
# TODO: Mount client_secret.json and credentials into a container somehow
# PYTHONPATH=. python vendor/youtube_upload/bin/youtube_upload --title=test --privacy=unlisted test.mp4
# TODO: Remove temp_dir if TEMP
return video_id
def save_hdf5_recordings_to_png(combine_all=False, sess_dir=None):
if combine_all:
hdf5_filenames = sorted(glob.glob(c.RECORDING_DIR + '/**/*.hdf5',
recursive=True))
else:
sess_dir = sess_dir or c.HDF5_SESSION_DIR
hdf5_filenames = sorted(glob.glob(sess_dir + '/*.hdf5', recursive=True))
save_dir = os.path.join(c.RECORDING_DIR, 'pngs', c.DATE_STR)
os.makedirs(save_dir)
for i, f in enumerate(hdf5_filenames):
try:
read_hdf5(f,
save_png_dir=save_dir,
save_prefix='hdf5_%s' % str(i).zfill(c.HDF5_DIR_ZFILL))
except OSError as e:
log.error(e)
return save_dir
def save_random_hdf5_to_png(recording_dir=c.RECORDING_DIR):
random_file = np.random.choice(glob.glob(recording_dir + '/*/*.hdf5'))
if not random_file:
raise RuntimeError('No hdf5 files found')
else:
p = os.path
save_png_dir = os.path.join(p.join(recording_dir, 'random_hdf5_view'),
p.basename(p.dirname(random_file)),
p.basename(random_file)[:-5])
log.info('Saving random files to ' + save_png_dir)
os.makedirs(save_png_dir, exist_ok=True)
read_hdf5(p.join(recording_dir, random_file),
save_png_dir=save_png_dir)
def read_hdf5_manual(recording_dir=c.RECORDING_DIR):
save_png_dir = os.path.join(recording_dir, 'test_view')
os.makedirs(save_png_dir, exist_ok=True)
read_hdf5(os.path.join(recording_dir, '2018-01-18__05-14-48PM',
'0000000001.hdf5'), save_png_dir=save_png_dir)
def is_debugging():
for frame in inspect.stack():
if frame[1].endswith("pydevd.py"):
return True
return False
def download_weights(url):
folder = url.split('/')[-1].replace('.zip', '')
dest = os.path.join(c.WEIGHTS_DIR, folder)
if not glob.glob(dest + '/*'):
log.info('Downloading weights %s', folder)
download(url, dest)
else:
log.info('Found cached weights at %s', dest)
return dest
def is_docker():
path = '/proc/self/cgroup'
return (
os.path.exists('/.dockerenv') or
os.path.isfile(path) and any('docker' in line for line in open(path))
)
def get_free_space_mb(filename):
"""Return folder/drive free space (in megabytes)."""
if platform.system() == 'Windows':
drive, _path = os.path.splitdrive(filename)
free_bytes = ctypes.c_ulonglong(0)
ctypes.windll.kernel32.GetDiskFreeSpaceExW(
ctypes.c_wchar_p(drive), None, None, ctypes.pointer(free_bytes))
return free_bytes.value / 1024 / 1024
else:
path = filename
while not os.path.exists(path):
if not path or path == '/':
raise ValueError('Drive does not exist for filename %s' % filename)
path = os.path.dirname(path)
st = os.statvfs(path)
return st.f_bavail * st.f_frsize / 1024 / 1024
def remotable(f):
def extract_args(*args, **kwargs):
return f((args, kwargs), *args, **kwargs)
return extract_args
def assert_disk_space(filename, mb=2000):
try:
if get_free_space_mb(filename) < mb:
raise Exception('Less than %dMB left on device, aborting'
' save of %s' % (mb, filename))
except Exception as e:
log.error('Could not get free space on the drive containing %s' %
filename)
raise e
def resize_images(input_image_shape, images, always=False):
import scipy.misc
for img_idx, img in enumerate(images):
img = images[img_idx]
if img.shape != input_image_shape or always:
# Interesting bug here. Since resize converts mean subtracted
# floats (~-120 to ~130) to 0-255 uint8,
# but we don't always resize since randomize_cameras does nothing
# to the size 5% of the time.
# This actually worked surprisingly well. Need to test whether
# this bug actually improves things or not.
log.debug('invalid image shape %s - resizing', str(img.shape))
images[img_idx] = scipy.misc.imresize(img, (input_image_shape[0],
input_image_shape[1]))
return images
def kill_process(process_to_kill):
try:
process_to_kill.terminate()
time.sleep(0.2)
i = 0
while process_to_kill and process_to_kill.poll() is None:
log.info('Waiting for process to die')
time.sleep(0.1 * 2 ** i)
if i > 4:
# Die!
log.warn('Forcefully killing process')
process_to_kill.kill()
return False
i += 1
return True
except Exception as e:
log.error('Error closing process', str(e))
return False
def get_valid_filename(s):
s = str(s).strip().replace(' ', '_')
return re.sub(r'(?u)[^-\w.]', '', s)
def copy_dir_clean(src, dest):
if exists(dest):
log.info('Removing %s', dest)
guarded_rmtree(dest)
log.info('Copying files to %s', dest)
shutil.copytree(src, dest)
def guarded_rmtree(dest, allow_small_paths=False):
msg = 'Not letting you delete %s' % dest
if dest in ['/', '\\', '/root', '~', expanduser('~')]:
raise RuntimeError(msg)
elif len(dest.split(os.path.sep)) <= 2:
raise RuntimeError(msg)
elif in_home(dest) and len(dest) < 10 and not allow_small_paths:
raise RuntimeError(msg)
elif not in_home(dest) and len(dest) < 5 and not allow_small_paths:
raise RuntimeError(msg)
else:
return shutil.rmtree(dest)
def sizeof_fmt(num, suffix='B'):
for unit in ['','Ki','Mi','Gi','Ti','Pi','Ei','Zi']:
if abs(num) < 1024.0:
return "%3.1f%s%s" % (num, unit, suffix)
num /= 1024.0
return "%.1f%s%s" % (num, 'Yi', suffix)
class timer:
def __init__(self, msg, fmt="%0.3g"):
self.msg = msg
self.fmt = fmt
def __enter__(self):
self.start = time.process_time()
return self
def __exit__(self, *args):
t = time.process_time() - self.start
log.debug(("%s : " + self.fmt + " seconds") % (self.msg, t))
self.time = t
def nearest_neighbor(me: np.array, them: np.array) -> Tuple[float, int]:
"""
:param me: start point
:param them: other points
:return:
"""
import scipy.spatial
if len(them) == 0:
distance, index = math.inf, -1
elif len(them) > 27:
# 27 derived from https://gist.github.com/crizCraig/fd3d04e28defa5d5cfc7f37757f81a26
with timer('kd tree total'):
kd_tree = scipy.spatial.cKDTree(them)
distance, index = kd_tree.query(me)
else:
with timer('brute force'):
min_dist = math.inf
min_index = -1
for index, point in enumerate(them):
dist = scipy.spatial.distance.cdist(
np.array([point]), np.array([me]))[0][0]
if dist < min_dist:
min_dist = dist
min_index = index
distance = min_dist
index = min_index
return distance, index
def dbox(obj=None, **kwargs):
if kwargs:
obj = dict(kwargs)
else:
obj = obj or {}
return Box(obj, default_box=True)
def main():
# download('https://d1y4edi1yk5yok.cloudfront.net/sim/asdf.zip', r'C:\Users\a\src\beta\deepdrive-agents-beta\asdf')
# read_hdf5_manual()
# ensure_sim()
# save_random_hdf5_to_png()
# assert_disk_space(r'C:\Users\a\DeepDrive\recordings\2018-11-03__12-29-33PM\0000000143.hdf5')
# assert_disk_space('/media/a/data-ext4/deepdrive-data/v2.1/asdf.hd5f')
# print(get_sim_url())
# print(save_recordings_to_png_and_mp4(png_dir='/tmp/tmp30zl8ouq'))
# print(save_hdf5_recordings_to_png())
# print(upload_to_gist('asdf', ['/home/c2/src/deepdrive/results/2018-05-30__02-40-01PM.csv', '/home/c2/src/deepdrive/results/2019-03-14__06-08-38PM.diff']))
# log.info('testing %s', os.path.expanduser('~'))
# import traceback
#
# traceback.print_stack(file=sys.stdout)
# log.info('testing %d', 1234)
# ensure_sim()
import scipy.spatial
# vehicle_positions = [[-24164.44140625, 29898.431640625, 19338.546875],
# [-21652.052734375, 32871.953125, 19180.44921875],
# [-18885.00390625, 35396.0, 19488.966796875],
# [-15975.9833984375, 37886.96875, 19873.97265625],
# [-6273.73681640625, 38792.83984375, 21070.3828125]]
vehicle_positions = np.random.normal(size=(27, 3))
with timer('kd tree total'):
with timer('kd tree build'):
kd_tree = scipy.spatial.cKDTree(vehicle_positions)
ego_position = [8000.45654297, 27470.125, 24762.65039062]
with timer('kd tree query'):
distance, index = kd_tree.query(ego_position)
with timer('brute force'):
min_dist = None
min_index = -1
for index, point in enumerate(vehicle_positions):
dist = scipy.spatial.distance.cdist(
np.array([point]), np.array([ego_position]))
if min_dist is None or dist < min_dist:
min_dist = dist
min_index = index
distance = min_dist
index = min_index
pass
if __name__ == '__main__':
main()