-
Notifications
You must be signed in to change notification settings - Fork 0
/
video_to_crop_ultralytics.py
206 lines (149 loc) · 8.11 KB
/
video_to_crop_ultralytics.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
# Be careful of using CVAT annotations with the original video as their height and width is half the original one.
from docopt import docopt
from contextlib import contextmanager
import cv2 as cv
import numpy as np
import os
import shutil
import sys
from utils.random_centered_crop import propose_random_crop, find_intersections, improve_crop, desired_movments
@contextmanager
def VideoCapture(input_video):
# findFileOrKeep allows more searching paths
capture = cv.VideoCapture(cv.samples.findFileOrKeep(input_video))
if not capture.isOpened():
print('Unable to open: ' + input_video, file=sys.stderr)
exit(0)
try:
yield capture
finally:
# Release the video capture object at the end
capture.release()
class PrecomputedMOTTracker():
def __init__(self, seq_path=None, verbose=False):
self.seq_dets = np.loadtxt(seq_path, delimiter=',')
self.last_frame = int(self.seq_dets[:, 0].max())
self.verbose = verbose
self.frames_seen = 0
def __call__(self, frame):
self.frames_seen = (self.frames_seen + 1) % 500
if self.verbose and self.frames_seen == 0:
print (f'Processing frame {frame}', file=sys.stderr)
tracks = self.seq_dets[self.seq_dets[:, 0] == frame, :]
return tracks
def random_crop(tracks, seen, crop_width, crop_height, width, height):
current = np.argwhere(~seen).flatten()[0]
trk = tracks[current]
initial, final, low, high = propose_random_crop(trk, crop_width, crop_height, width, height)
intersect = find_intersections(tracks, initial, final)
unseen_intersect = intersect & (~seen)
unseen_tracks = tracks[unseen_intersect, :]
initial, final = improve_crop(unseen_tracks, initial, final, low, high, width, height)
return initial, final
def adjust_annotations(tracks, seen, initial, final, crop_width, crop_height):
delta_w, delta_h = desired_movments(tracks, initial, final)
seen = seen | ((delta_w == 0) & (delta_h == 0))
within = ((np.abs(delta_w) < tracks[:, 4]) & (np.abs(delta_h) < tracks[:, 5]))
tracks_save = tracks[within, :].copy()
new_left = tracks_save[:, 2] - initial[0]
tracks_save[:, 2] = np.clip(new_left, 0, None)
tracks_save[:, 4] = np.minimum(tracks_save[:, 4], tracks_save[:, 4] + new_left)
tracks_save[:, 4] = np.minimum(tracks_save[:, 4], crop_width - tracks_save[:, 2])
new_up = tracks_save[:, 3] - initial[1]
tracks_save[:, 3] = np.clip(new_up, 0, None)
tracks_save[:, 5] = np.minimum(tracks_save[:, 5], tracks_save[:, 5] + new_up)
tracks_save[:, 5] = np.minimum(tracks_save[:, 5], crop_height - tracks_save[:, 3])
return tracks_save, seen
def process_video(video_path, seq_path, sampling_rate, test_frac, crop_width, crop_height, basename, val_img_dir, val_label_dir, train_img_dir, train_label_dir, video_id, verbose=True):
tracker = PrecomputedMOTTracker(seq_path, verbose=verbose)
#save_frames = np.arange(1, tracker.last_frame, sampling_rate, dtype=int)
save_frames = np.unique(tracker.seq_dets[:, 0])[::sampling_rate].astype(int)
valid_frames = save_frames.copy()
#np.random.shuffle(valid_frames)
valid_frames = valid_frames[: int(len(valid_frames) * test_frac)]
with VideoCapture(video_path) as capture:
width = capture.get(cv.CAP_PROP_FRAME_WIDTH)
height = capture.get(cv.CAP_PROP_FRAME_HEIGHT)
if crop_width > width or crop_height > height:
print(f'crop_width = {crop_width} >? width = {width}\ncrop_height = {crop_height} >? height = {height}')
raise Exception(f'crop_width = {crop_width} >? width = {width}\ncrop_height = {crop_height} >? height = {height}')
for fr in save_frames:
tracks = tracker(fr)
tracks[:, 2] = np.maximum(tracks[:, 2], 0)
tracks[:, 3] = np.maximum(tracks[:, 3], 0)
tracks[:, 4] = np.minimum(tracks[:, 4], width - tracks[:, 2])
tracks[:, 5] = np.minimum(tracks[:, 5], height - tracks[:, 3])
capture.set(cv.CAP_PROP_POS_FRAMES, fr - 1)
_, frame = capture.read()
if frame is None:
print (f'Frame {fr} is None')
break
# Crop and each uncentered and uncut ant group is in one and only one output!
seen = np.full(len(tracks), False)
seen[tracks[:, 4] > crop_width] = True
seen[tracks[:, 5] > crop_height] = True
idx = 1
while not np.all(seen):
initial, final = random_crop(tracks, seen, crop_width, crop_height, width, height)
img = frame[initial[1] : final[1], initial[0] : final[0]]
tracks_save, seen = adjust_annotations(tracks, seen, initial, final, crop_width, crop_height)
base_filename = f'{basename}_{video_id}_{fr:06}_{idx}_{len(tracks_save)}'
filename = f'{base_filename}.png'
labels_filename = f'{base_filename}.txt'
idx += 1
mot2yolo = lambda trk : ['0', f'{(trk[2] + (trk[4] / 2)) / crop_width}', f'{(trk[3] + (trk[5] / 2)) / crop_height}', f'{trk[4] / crop_width}', f'{trk[5] / crop_height}']
labels = '\n'.join([' '.join(mot2yolo(trk)) for trk in tracks_save])
if fr in valid_frames:
cv.imwrite(os.path.join(val_img_dir, filename), img)
with open(os.path.join(val_label_dir, labels_filename), 'w') as f:
f.write(labels)
else:
cv.imwrite(os.path.join(train_img_dir, filename), img)
with open(os.path.join(train_label_dir, labels_filename), 'w') as f:
f.write(labels)
DOCTEXT = f"""
Usage:
video_to_crop_ultralytics.py <output_file> (<video_path> <seq_path>)... [--test_frac=<tf>] [--sampling_rate=<sr>] [--width=<w>] [--height=<h>]
Options:
--test_frac=<tf> The fraction of frames used for testing. [default: 0.3]
--sampling_rate=<sr> Number of frames skipped between saved images. [default: 2]
--width=<w> Width of the crop. [default: 640]
--height=<h> Height of the crop. [default: 640]
"""
if __name__ == "__main__":
args = docopt(DOCTEXT, argv=sys.argv[1:], help=True, version=None, options_first=False)
video_pathes = args['<video_path>']
seq_pathes = args['<seq_path>']
output_file = args['<output_file>']
test_frac = float(args['--test_frac'])
sampling_rate = int(args['--sampling_rate'])
crop_width = int(args['--width'])
crop_height = int(args['--height'])
basename = os.path.basename(output_file)
yolo_config_dir = os.path.join(output_file, basename)
train_img_dir = os.path.join(output_file, basename, 'images', 'train')
val_img_dir = os.path.join(output_file, basename, 'images', 'val')
train_label_dir = os.path.join(output_file, basename, 'labels', 'train')
val_label_dir = os.path.join(output_file, basename, 'labels', 'val')
os.makedirs(yolo_config_dir, exist_ok=False)
os.makedirs(train_img_dir, exist_ok=False)
os.makedirs(val_img_dir, exist_ok=False)
os.makedirs(train_label_dir, exist_ok=False)
os.makedirs(val_label_dir, exist_ok=False)
for i, (video_path, seq_path) in enumerate(zip(video_pathes, seq_pathes)):
print(f'VIDEO {i + 1} OF {len(video_pathes)}')
process_video(video_path, seq_path, sampling_rate, test_frac, crop_width, crop_height, basename, val_img_dir, val_label_dir, train_img_dir, train_label_dir, i, verbose=True)
config_text = f"""
# Train/val/test sets as 1) dir: path/to/imgs, 2) file: path/to/imgs.txt, or 3) list: [path/to/imgs1, path/to/imgs2, ..]
path: ./{output_file} # dataset root dir
train: images/train # train images (relative to 'path') 128 images
val: images/val # val images (relative to 'path') 128 images
test: # test images (optional)
# Classes
names:
0: ant
"""
with open(os.path.join(yolo_config_dir, f'{basename}.yaml'), 'w') as f:
f.write(config_text)
shutil.make_archive(output_file, 'zip', output_file)
shutil.rmtree(output_file)