mmaction2
mmaction2 copied to clipboard
Published
20 hours ago •
open-mmlab
open-mmlab
Run "demo/webcam_demo_spatiotemporal_det.py" , the program is stuck.
When I run "demo/webcam_demo_spatiotemporal_det.py" with USB cam or RTSP web cam , the program is stuck after running for a period of time.
I think this is caused by the inference speed or the display speed being slower than the reading speed.
Some log information is as follows:（Sometimes directly show killed）
INFO:__main__:Stdet Results: [[('sit', 0.8451391), ('carry/hold (an object)', 0.52010876)]]
DEBUG:__main__:Main thread inference time 627 ms
DEBUG:__main__:Read thread: 691 ms, 12 fps
DEBUG:__main__:Read thread: 231 ms, 35 fps
INFO:__main__:Stdet Results: [[('sit', 0.84119666), ('carry/hold (an object)', 0.5008541)]]
DEBUG:__main__:Main thread inference time 600 ms
DEBUG:__main__:Read thread: 296 ms, 27 fps
+++++++++++++++++++++++++++++++++++++++++++++++++++++++
DEBUG:__main__:Read thread: 212 ms, 38 fps
DEBUG:__main__:Read thread: 332 ms, 24 fps
DEBUG:__main__:Read thread: 350 ms, 23 fps
+++++++++++++++++++++++++++++++++++++++++++++++++++++++
INFO:__main__:Stdet Results: [[('sit', 0.8407789), ('carry/hold (an object)', 0.48366198)]]
DEBUG:__main__:Main thread inference time 1010 ms
DEBUG:__main__:Read thread: 199 ms, 40 fps
+++++++++++++++++++++++++++++++++++++++++++++++++++++++
DEBUG:__main__:Read thread: 320 ms, 25 fps
INFO:__main__:Stdet Results: [[('sit', 0.8350647), ('carry/hold (an object)', 0.46698186), ('text on/look at a cellphone', 0.42576995)]]
DEBUG:__main__:Main thread inference time 657 ms
DEBUG:__main__:Read thread: 250 ms, 32 fps
DEBUG:__main__:Read thread: 247 ms, 32 fps
+++++++++++++++++++++++++++++++++++++++++++++++++++++++
INFO:__main__:Stdet Results: [[('sit', 0.80609334), ('carry/hold (an object)', 0.48520392), ('text on/look at a cellphone', 0.5238202)]]
DEBUG:__main__:Main thread inference time 382 ms
DEBUG:__main__:Read thread: 251 ms, 32 fps
DEBUG:__main__:Read thread: 316 ms, 25 fps
DEBUG:__main__:Read thread: 273 ms, 29 fps
INFO:__main__:Stdet Results: [[('sit', 0.7361066), ('carry/hold (an object)', 0.47983062)]]
DEBUG:__main__:Main thread inference time 957 ms
+++++++++++++++++++++++++++++++++++++++++++++++++++++++
DEBUG:__main__:Read thread: 256 ms, 31 fps
DEBUG:__main__:Read thread: 243 ms, 33 fps
DEBUG:__main__:Read thread: 300 ms, 27 fps
DEBUG:__main__:Read thread: 273 ms, 29 fps
+++++++++++++++++++++++++++++++++++++++++++++++++++++++
INFO:__main__:Stdet Results: [[('sit', 0.7959219), ('carry/hold (an object)', 0.53366905)]]
DEBUG:__main__:Main thread inference time 970 ms
DEBUG:__main__:Read thread: 208 ms, 39 fps
DEBUG:__main__:Read thread: 350 ms, 23 fps
DEBUG:__main__:Read thread: 238 ms, 34 fps
INFO:__main__:Stdet Results: [[('sit', 0.82904106), ('carry/hold (an object)', 0.60257)]]
DEBUG:__main__:Main thread inference time 813 ms
+++++++++++++++++++++++++++++++++++++++++++++++++++++++
DEBUG:__main__:Read thread: 217 ms, 37 fps
DEBUG:__main__:Read thread: 318 ms, 25 fps
DEBUG:__main__:Read thread: 332 ms, 24 fps
INFO:__main__:Stdet Results: [[('sit', 0.8334102), ('carry/hold (an object)', 0.6356909)]]
DEBUG:__main__:Main thread inference time 1003 ms
+++++++++++++++++++++++++++++++++++++++++++++++++++++++
DEBUG:__main__:Read thread: 312 ms, 26 fps
DEBUG:__main__:Read thread: 228 ms, 35 fps
+++++++++++++++++++++++++++++++++++++++++++++++++++++++
DEBUG:__main__:Read thread: 1157 ms, 7 fps
INFO:__main__:Stdet Results: [[('sit', 0.8563365), ('carry/hold (an object)', 0.6350206)]]
DEBUG:__main__:Main thread inference time 8273 ms
DEBUG:__main__:Read thread: 11494 ms, 1 fps
Process finished with exit code 137 (interrupted by signal 9: SIGKILL)
I have adjust the keyframes:
keyframe = task.frames[len(task.frames) // 4]
However, it will still get stuck.
What parameters should I adjust so that the log information output will not get stuck.
Parameter setting:
def parse_args():
parser = argparse.ArgumentParser(
description='MMAction2 webcam spatio-temporal detection demo')
parser.add_argument(
'--config',
default=('../configs/detection/ava/'
'slowonly_omnisource_pretrained_r101_8x8x1_20e_ava_rgb.py'),
help='spatio temporal detection config file path')
parser.add_argument(
'--checkpoint',
default=('https://download.openmmlab.com/mmaction/detection/ava/'
'slowonly_omnisource_pretrained_r101_8x8x1_20e_ava_rgb/'
'slowonly_omnisource_pretrained_r101_8x8x1_20e_ava_rgb'
'_20201217-16378594.pth'),
help='spatio temporal detection checkpoint file/url')
parser.add_argument(
'--action-score-thr',
type=float,
default=0.4,
help='the threshold of human action score')
parser.add_argument(
'--det-config',
default='../demo/faster_rcnn_r50_fpn_2x_coco.py',
help='human detection config file path (from mmdet)')
parser.add_argument(
'--det-checkpoint',
default=('http://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/'
'faster_rcnn_r50_fpn_2x_coco/'
'faster_rcnn_r50_fpn_2x_coco_'
'bbox_mAP-0.384_20200504_210434-a5d8aa15.pth'),
help='human detection checkpoint file/url')
parser.add_argument(
'--det-score-thr',
type=float,
default=0.9,
help='the threshold of human detection score')
parser.add_argument(
'--input-video',
default='0',
type=str,
help='webcam id or input video file/url')
parser.add_argument(
'--label-map', default='../demo/label_map_ava.txt', help='label map file')
parser.add_argument(
'--device', type=str, default='cuda:0', help='CPU/CUDA device option')
parser.add_argument(
'--output-fps',
default=15,
type=int,
help='the fps of demo video output')
parser.add_argument(
'--out-filename',
default=None,
type=str,
help='the filename of output video')
parser.add_argument(
'--show',
# action='store_true',
default=True,
help='Whether to show results with cv2.imshow')
parser.add_argument(
'--display-height',
type=int,
default=0,
help='Image height for human detector and draw frames.')
parser.add_argument(
'--display-width',
type=int,
default=0,
help='Image width for human detector and draw frames.')
parser.add_argument(
'--predict-stepsize',
default=8,
type=int,
help='give out a prediction per n frames')
parser.add_argument(
'--clip-vis-length',
default=8,
type=int,
help='Number of draw frames per clip.')
parser.add_argument(
'--cfg-options',
nargs='+',
action=DictAction,
default={},
help='override some settings in the used config, the key-value pair '
'in xxx=yyy format will be merged into config file. For example, '
"'--cfg-options model.backbone.depth=18 model.backbone.with_cp=True'")
args = parser.parse_args()
return args
solve it. adjust param.
Hello, how is this parameter adjusted? Do you change the parameters according to this? keyframe = task.frames[len(task.frames) // 4] Try changing the number 4, make it bigger?
solve it. adjust param.
Hello, how is this parameter adjusted? Do you change the parameters according to this? keyframe = task.frames[len(task.frames) // 4] Try changing the number 4, make it bigger?
I can't rememeber it. May adjust "--predict-stepsize --output-fps". No imshow.
solve it. adjust param.
Hello, how is this parameter adjusted? Do you change the parameters according to this? keyframe = task.frames[len(task.frames) // 4] Try changing the number 4, make it bigger?
I can't rememeber it. May adjust "--predict-stepsize --output-fps". No imshow.
OK, thank you. Is your camera used in conjunction with the GUI interface? Is there video behavior detection during camera detection?