-
Notifications
You must be signed in to change notification settings - Fork 5
/
test_video_ava.py
408 lines (328 loc) · 13.4 KB
/
test_video_ava.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
import argparse
import cv2
import os
import time
import numpy as np
import torch
from PIL import Image
from dataset.transforms import BaseTransform
from utils.misc import load_weight
from config import build_dataset_config, build_model_config
from models.detector import build_model
def parse_args():
parser = argparse.ArgumentParser(description='YOWOF')
# basic
parser.add_argument('-size', '--img_size', default=320, type=int,
help='the size of input frame')
parser.add_argument('--show', action='store_true', default=False,
help='show the visulization results.')
parser.add_argument('--cuda', action='store_true', default=False,
help='use cuda.')
parser.add_argument('--save_folder', default='det_results/', type=str,
help='Dir to save results')
parser.add_argument('-vs', '--vis_thresh', default=0.35, type=float,
help='threshold for visualization')
parser.add_argument('--video', default='9Y_l9NsnYE0.mp4', type=str,
help='AVA video name.')
parser.add_argument('-d', '--dataset', default='ava_v2.2',
help='ava_v2.2')
# model
parser.add_argument('-v', '--version', default='yowof-r18', type=str,
help='build yowof')
parser.add_argument('--weight', default=None, type=str,
help='Trained state_dict file path to open')
parser.add_argument('--topk', default=40, type=int,
help='NMS threshold')
parser.add_argument('-inf', '--inf_mode', default='clip', type=str, choices=['clip', 'semi_stream', 'stream'],
help='inference mode: clip or stream')
return parser.parse_args()
@torch.no_grad()
def detect_clip(args, d_cfg, model, device, transform, class_names):
# path to save
save_path = os.path.join(args.save_folder, 'ava_video')
os.makedirs(save_path, exist_ok=True)
# path to video
path_to_video = os.path.join(d_cfg['data_root'], 'videos_15min', args.video)
# video
video = cv2.VideoCapture(path_to_video)
fourcc = cv2.VideoWriter_fourcc(*'XVID')
save_size = (640, 480)
save_name = os.path.join(save_path, 'detection.avi')
fps = 15.0
out = cv2.VideoWriter(save_name, fourcc, fps, save_size)
# run
video_clip = []
while(True):
ret, frame = video.read()
if ret:
# to PIL image
frame_pil = Image.fromarray(frame.astype(np.uint8))
# prepare
if len(video_clip) <= 0:
for _ in range(d_cfg['len_clip']):
video_clip.append(frame_pil)
video_clip.append(frame_pil)
del video_clip[0]
# orig size
orig_h, orig_w = frame.shape[:2]
# transform
x, _ = transform(video_clip)
# List [T, 3, H, W] -> [T, 3, H, W]
x = torch.stack(x)
x = x.unsqueeze(0).to(device) # [B, T, 3, H, W], B=1
t0 = time.time()
# inference
out_bboxes = model(x)
out_bboxes = out_bboxes[0]
print("inference time ", time.time() - t0, "s")
# visualize detection results
for bbox in out_bboxes:
x1, y1, x2, y2 = bbox[:4]
cls_out = bbox[4:]
# rescale bbox
x1, x2 = int(x1 * orig_w), int(x2 * orig_w)
y1, y2 = int(y1 * orig_h), int(y2 * orig_h)
cls_scores = np.array(cls_out)
indices = np.where(cls_scores > args.vis_thresh)
scores = cls_scores[indices]
indices = list(indices[0])
scores = list(scores)
cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 255, 0), 2)
if len(scores) > 0:
blk = np.zeros(frame.shape, np.uint8)
font = cv2.FONT_HERSHEY_SIMPLEX
coord = []
text = []
text_size = []
for _, cls_ind in enumerate(indices):
text.append("[{:.2f}] ".format(scores[_]) + str(class_names[cls_ind]))
text_size.append(cv2.getTextSize(text[-1], font, fontScale=0.25, thickness=1)[0])
coord.append((x1+3, y1+7+10*_))
cv2.rectangle(blk, (coord[-1][0]-1, coord[-1][1]-6), (coord[-1][0]+text_size[-1][0]+1, coord[-1][1]+text_size[-1][1]-4), (0, 255, 0), cv2.FILLED)
frame = cv2.addWeighted(frame, 1.0, blk, 0.25, 1)
for t in range(len(text)):
cv2.putText(frame, text[t], coord[t], font, 0.25, (0, 0, 0), 1)
# save
out.write(frame)
if args.show:
# show
cv2.imshow('key-frame detection', frame)
cv2.waitKey(1)
else:
break
video.release()
out.release()
cv2.destroyAllWindows()
@torch.no_grad()
def detect_stream(args, d_cfg, model, device, transform, class_names):
# path to save
save_path = os.path.join(args.save_folder, 'ava_video')
os.makedirs(save_path, exist_ok=True)
# path to video
path_to_video = os.path.join(d_cfg['data_root'], 'videos_15min', args.video)
# video
video = cv2.VideoCapture(path_to_video)
fourcc = cv2.VideoWriter_fourcc(*'XVID')
save_size = (640, 480)
save_name = os.path.join(save_path, 'detection.avi')
fps = 15.0
out = cv2.VideoWriter(save_name, fourcc, fps, save_size)
# initalize model
model.initialization = True
# run
video_clip = []
while(True):
ret, frame = video.read()
if ret:
# to PIL image
frame_pil = Image.fromarray(frame.astype(np.uint8))
# prepare
if len(video_clip) <= 0:
for _ in range(d_cfg['len_clip']):
video_clip.append(frame_pil)
video_clip.append(frame_pil)
del video_clip[0]
# orig size
orig_h, orig_w = frame.shape[:2]
# transform
x, _ = transform(video_clip)
# List [T, 3, H, W] -> [T, 3, H, W]
x = torch.stack(x)
x = x.unsqueeze(0).to(device) # [B, T, 3, H, W], B=1
t0 = time.time()
# inference
out_bboxes = model(x)
print("inference time ", time.time() - t0, "s")
# visualize detection results
for bbox in out_bboxes:
x1, y1, x2, y2 = bbox[:4]
cls_out = bbox[4:]
# rescale bbox
x1, x2 = int(x1 * orig_w), int(x2 * orig_w)
y1, y2 = int(y1 * orig_h), int(y2 * orig_h)
cls_scores = np.array(cls_out)
indices = np.where(cls_scores > args.vis_thresh)
scores = cls_scores[indices]
indices = list(indices[0])
scores = list(scores)
cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 255, 0), 2)
if len(scores) > 0:
blk = np.zeros(key_frame.shape, np.uint8)
font = cv2.LINE_AA
coord = []
text = []
text_size = []
# scores, indices = [list(a) for a in zip(*sorted(zip(scores,indices), reverse=True))] # if you want, you can sort according to confidence level
for _, cls_ind in enumerate(indices):
text.append("[{:.2f}] ".format(scores[_]) + str(class_names[cls_ind]))
text_size.append(cv2.getTextSize(text[-1], font, fontScale=0.5, thickness=1)[0])
coord.append((x1+3, y1+14+20*_))
cv2.rectangle(blk, (coord[-1][0]-1, coord[-1][1]-12), (coord[-1][0]+text_size[-1][0]+1, coord[-1][1]+text_size[-1][1]-4), (0, 255, 0), cv2.FILLED)
key_frame = cv2.addWeighted(key_frame, 1.0, blk, 0.5, 1)
for t in range(len(text)):
cv2.putText(key_frame, text[t], coord[t], font, 0.5, (0, 0, 0), 1)
# save
out.write(frame)
if args.show:
# show
cv2.imshow('key-frame detection', frame)
cv2.waitKey(1)
else:
break
video.release()
out.release()
cv2.destroyAllWindows()
@torch.no_grad()
def detect_semi_stream(args, d_cfg, model, device, transform, class_names):
# path to save
save_path = os.path.join(args.save_folder, 'ava_video')
os.makedirs(save_path, exist_ok=True)
# path to video
path_to_video = os.path.join(d_cfg['data_root'], 'videos_15min', args.video)
# video
video = cv2.VideoCapture(path_to_video)
fourcc = cv2.VideoWriter_fourcc(*'XVID')
save_size = (640, 480)
save_name = os.path.join(save_path, 'detection.avi')
fps = 15.0
out = cv2.VideoWriter(save_name, fourcc, fps, save_size)
# run
video_clip = []
while(True):
ret, frame = video.read()
if ret:
# to PIL image
frame_pil = Image.fromarray(frame.astype(np.uint8))
# prepare
if len(video_clip) <= 0:
for _ in range(d_cfg['len_clip']):
video_clip.append(frame_pil)
video_clip.append(frame_pil)
del video_clip[0]
# orig size
orig_h, orig_w = frame.shape[:2]
# transform
x, _ = transform(video_clip)
# List [T, 3, H, W] -> [T, 3, H, W]
x = torch.stack(x)
x = x.unsqueeze(0).to(device) # [B, T, 3, H, W], B=1
t0 = time.time()
# inference
out_bboxes = model(x)
print("inference time ", time.time() - t0, "s")
# visualize detection results
for bbox in out_bboxes:
x1, y1, x2, y2 = bbox[:4]
cls_out = bbox[4:]
# rescale bbox
x1, x2 = int(x1 * orig_w), int(x2 * orig_w)
y1, y2 = int(y1 * orig_h), int(y2 * orig_h)
cls_scores = np.array(cls_out)
indices = np.where(cls_scores > args.vis_thresh)
scores = cls_scores[indices]
indices = list(indices[0])
scores = list(scores)
cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 255, 0), 2)
if len(scores) > 0:
blk = np.zeros(frame.shape, np.uint8)
font = cv2.FONT_HERSHEY_SIMPLEX
coord = []
text = []
text_size = []
for _, cls_ind in enumerate(indices):
text.append("[{:.2f}] ".format(scores[_]) + str(class_names[cls_ind]))
text_size.append(cv2.getTextSize(text[-1], font, fontScale=0.25, thickness=1)[0])
coord.append((x1+3, y1+7+10*_))
cv2.rectangle(blk, (coord[-1][0]-1, coord[-1][1]-6), (coord[-1][0]+text_size[-1][0]+1, coord[-1][1]+text_size[-1][1]-4), (0, 255, 0), cv2.FILLED)
frame = cv2.addWeighted(frame, 1.0, blk, 0.25, 1)
for t in range(len(text)):
cv2.putText(frame, text[t], coord[t], font, 0.25, (0, 0, 0), 1)
# save
out.write(frame)
if args.show:
# show
cv2.imshow('key-frame detection', frame)
cv2.waitKey(1)
else:
break
video.release()
out.release()
cv2.destroyAllWindows()
if __name__ == '__main__':
args = parse_args()
# cuda
if args.cuda:
print('use cuda')
device = torch.device("cuda")
else:
device = torch.device("cpu")
# config
d_cfg = build_dataset_config(args)
m_cfg = build_model_config(args)
class_names = d_cfg['label_map']
num_classes = 80
# transform
basetransform = BaseTransform(
img_size=d_cfg['test_size'],
pixel_mean=d_cfg['pixel_mean'],
pixel_std=d_cfg['pixel_std']
)
# build model
model = build_model(
args=args,
d_cfg=d_cfg,
m_cfg=m_cfg,
device=device,
num_classes=num_classes,
trainable=False
)
# load trained weight
model = load_weight(model=model, path_to_ckpt=args.weight)
# to eval
model = model.to(device).eval()
# inference mode
model.set_inference_mode(args.inf_mode)
if args.inf_mode == 'clip':
# run
detect_clip(args=args,
d_cfg=d_cfg,
model=model,
device=device,
transform=basetransform,
class_names=class_names)
elif args.inf_mode == 'stream':
# run
detect_stream(args=args,
d_cfg=d_cfg,
model=model,
device=device,
transform=basetransform,
class_names=class_names)
elif args.inf_mode == 'semi_stream':
# run
detect_semi_stream(args=args,
d_cfg=d_cfg,
model=model,
device=device,
transform=basetransform,
class_names=class_names)