-
Notifications
You must be signed in to change notification settings - Fork 0
/
gpu_flow.py
282 lines (220 loc) · 8.91 KB
/
gpu_flow.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
#!/usr/bin/python3
# Imports
import os
import argparse
import time
#import tqdm
import cv2
import numpy as np
# Functions
def to_img(raw_flow, min_bound, max_bound,
pix_min, pix_max):
"""
Convert raw flow data from min-max bound to pixel bounds.
"""
# First clip extreme values
clipped_flow = raw_flow.clip(min_bound, max_bound)
# Then scale to [0, 1]
zero_one_flow = np.true_divide(clipped_flow - min_bound, max_bound - min_bound)
# And finally convert to new range
pix_flow = zero_one_flow * (pix_max - pix_min) + pix_min
return pix_flow.astype(np.uint8)
def read_video(video_file):
"""
Read a video file from disk, returning cv2 video capture object.
WARNING: the capture needs to be .release()'d in every possible
code execution path.
"""
cap = cv2.VideoCapture(video_file)
if not cap.isOpened():
print('Error opening video stream: {}' \
.format(video_file))
cap.release()
raise IOError
print('Opened video capture, total frames: {}' \
.format(cap.get(cv2.CAP_PROP_FRAME_COUNT)))
return cap
def save_flows(flow, save_path, frame_num, bounds, pixrange, xmark='x', ymark='y', filetype='png'):
"""
Function that saves flow images to disk after postprocessing.
"""
args = bounds + pixrange
print('Flow shape: {}'.format(flow.shape))
flow_img = to_img(flow, *args)
#tarshape = list(flow_img[..., 0].shape)
#tarshape.append(3)
#target_x = np.zeros(tuple(tarshape))
#target_y = np.zeros(tuple(tarshape))
#for i in range(3):
# target_x[..., i] = flow_img[..., 0]
# target_y[..., i] = flow_img[..., 1]
#target_x = to_img(flow[..., 0], *args)
#target_y = to_img(flow[..., 1], *args)
target_x = flow_img[..., 0]
target_y = flow_img[..., 1]
print('DEBUG flow images')
print(type(target_x))
print(target_x.shape)
print(target_x.dtype)
fname = 'flow_{}'+'_{:08d}.{}'.format(frame_num, filetype)
print(os.path.join(save_path, fname.format('??')))
#cv2.imwrite('temp.png', target_x)#, [int(cv2.IMWRITE_JPEG_QUALITY), 90])
cv2.imwrite(os.path.join(save_path, fname.format(xmark)), target_x)#, [int(cv2.IMWRITE_JPEG_QUALITY), 90])
cv2.imwrite(os.path.join(save_path, fname.format(ymark)), target_y)#, [int(cv2.IMWRITE_JPEG_QUALITY), 90])
def comp_next_flow(video_capture, prev_gray, flow_comp, \
frame_num, bounds, pixrange, save_path, use_cuda, \
filetype='png', save_img=False):
"""
Get next optical flow frame, given video capture object at the
current frame and the preious camera frame.
"""
# Get next camera frame
flag, new_frame = video_capture.read()
# Check if the video is over
if not flag:
return None
# check if the raw frames need to be saved
if save_img:
cv2.imwrite(new_frame, os.path.join(save_path, \
'frame_{:08d}.{}'.format(frame_num, filetype)))
# Compute auxiliary variables
gray_raw = cv2.cvtColor(new_frame, cv2.COLOR_RGB2GRAY)
if (use_cuda):
gray = cv2.cuda_GpuMat()
#prev_gray = cv2.cuda_GpuMat()
gray.upload(gray_raw)
#prev_gray.upload(prev_gray)
else:
gray = gray_raw
print('Computing flow!')
# Compute flow
flow_raw = flow_comp(prev_gray, gray, None)
if (use_cuda):
flow = flow_raw.download()
else:
flow = flow_raw
#save_flows(flow.image, save_path, frame_num, bounds, pixrange)
save_flows(flow, save_path, frame_num, bounds, pixrange)
return gray
def skip_frames(video_capture, num_frames):
"""
Advance the video_capture object by num_frames.
Return number of frames passed over.
"""
ret = 0
for t in range(num_frames):
flag, _ = video_capture.read()
if flag:
ret += 1
return ret
def flowify(video_file, step, bounds, save_path, no_cuda, pixrange=[0,255]):
"""
Extract flow images from an entire video.
"""
# Start the video reading
cap = read_video(video_file)
try:
cuda_avail = cv2.cuda.getCudaEnabledDeviceCount() > 0
except:
cuda_avail = False
use_cuda = cuda_avail and (not no_cuda)
loop_times = [] # keep some perf stats
try:
# Set up flow computation
#flowDTLV1 = cv2.createOptFlow_DualTVL1()
#flowDTLV1 = cv2.DualTVL1OpticalFlow_create()
if (use_cuda):
print('Setting CUDA optical flow computation...')
flowDTLV1 = cv2.cuda_OpticalFlowDual_TVL1.create()
else:
print('Setting CPU opticla flow computation...')
flowDTLV1 = cv2.optflow.DualTVL1OpticalFlow_create()
flow_comp = flowDTLV1.calc
# Bootstrap with the first frame
_, first_frame = cap.read()
first_gray_raw = cv2.cvtColor(first_frame, cv2.COLOR_RGB2GRAY)
if use_cuda:
first_gray = cv2.cuda_GpuMat()
first_gray.upload(first_gray_raw)
else:
first_gray = first_gray_raw
frame_num = 1
# Start and loop
frame_num += skip_frames(cap, step-1)
gray = comp_next_flow(cap, first_gray, flow_comp, \
frame_num, bounds, pixrange, save_path, use_cuda)
frame_num += 1
while (gray is not None):
start = time.time()
frame_num += skip_frames(cap, step-1)
print('DEBUG frame number')
print('expected: {}, actual: {}'.format(frame_num, \
cap.get(cv2.CAP_PROP_POS_FRAMES)))
new_gray = comp_next_flow(cap, gray, flow_comp, \
frame_num, bounds, pixrange, save_path, use_cuda)
gray = new_gray
frame_num += 1
end = time.time()
loop_times.append(end-start)
finally:
# always release video resource
cap.release()
print('Looping avg/std: {}, {}' \
.format(np.mean(loop_times), \
np.std(loop_times)))
def get_video_list(root_path, recursive=False, \
valid_exts=['.mp4', '.avi']):
"""
Get list of files to parse from a root directory.
"""
video_list = []
for sub in os.listdir(root_path):
sub_path = os.path.join(root_path, sub)
if (os.path.isdir(sub_path)):
if recursive:
video_list.append(get_video_list(sub_path))
elif (os.path.splitext(sub_path)[-1] in valid_exts):
video_list.append(sub_path)
return video_list
def execute(vids, step, bound, serial, save_path, no_cuda):
"""
Execution manager for extracting flow from a batch of videos.
If serial is False, multiprocessing is used, else videos are
processed one at a time.
"""
if serial:
for v in vids:
flowify(v, step, [-bound, bound], save_path, no_cuda)
else:
raise NotImplementedError
# Driver code
def main(args):
assert not (args.no_cuda and args.cuda_device), \
'Impossible input arguments configuration'
if args.cuda_device:
cv2.cuda.setDevice(args.cuda_device)
print('Finding videos in: {}'.format(args.data_root))
vl = get_video_list(args.data_root, args.recursive)
if (args.verbose):
print(vl)
print('Done, found {} in total.'.format(len(vl)))
print('Executing flow computations...')
execute(vl, args.frame_step, args.bound, args.serial, \
args.output_dir, args.no_cuda)
print('Done, results in: {}'.format(args.output_dir))
if __name__=='__main__':
parser = argparse.ArgumentParser(description='''
Extract dense optical flow from video files.
''')
parser.add_argument('data_root', help='root directory to look for data')
parser.add_argument('--recursive', '-r', action='store_true', help='add this switch to recursively search data_root for files, otherwise only looks at the direct contents of data_root only')
parser.add_argument('output_dir', help='output directory')
parser.add_argument('--serial', '-s', action='store_true', help='add this switch to disable multiprocessing')
parser.add_argument('--frame_step', '-f', type=int, default=1, help='determines the number of frames between flow computation data (default: 1)')
parser.add_argument('--bound', '-b', type=int, default=20, help='the absolute bound on fow magnitude, i.e. flow will be in: [-bound, bound] (default: 20)')
parser.add_argument('--verbose', '-v', action='store_true', help='add this switch for extra messages during execution')
parser.add_argument('--no_cuda', '-n', action='store_true', help='add this switch to force CPU execution')
parser.add_argument('--cuda_device', '-c', type=int, help='choose specific device to run on')
parser.set_defaults(serial=False, recursive=False, verbose=False, no_cuda=False)
args = parser.parse_args()
main(args)