FFMPEGCV incorrectly reads video fps as 60 most of the time about deep-license-plate-recognition

[akonanykhin]

Seems like fps_cap.get(cv2.CAP_PROP_FPS) only reports incorrect FPS for the version of OpenCV we're currently using. 4.7.0.72 does not have this problem, for example. So rolling OpenCV back (or forward) could solve this issue.

[danleyb2]

@akonanykhin Could you kindly share a test video of this? I have test several videos and unable to reproduce

[akonanykhin]

@danleyb2 It's one of those 2:
https://drive.google.com/file/d/11le_bXaoZZtr0-sBRRsED4ZFYd0mT8jk/view?usp=sharing
https://drive.google.com/file/d/1JiCurUGvTVczErST2nruJyqyBCOWa5yD/view?usp=sharing

[danleyb2]

@akonanykhin
Videos don't have a defined fps defined but there is a tbr that can be a replacement.
Upgrading OpenCV resolves the issue but existing version of ffmpegcv didn't have the issue either.

Fprobe output:

Input #0, matroska,webm, from '/home/danleyb2/Work/platerecognizer/stream-test-videos/trinam.mkv':
  Metadata:
    encoder         : MilestoneMux
  Duration: 00:01:59.87, start: 0.000000, bitrate: 2259 kb/s
  Stream #0:0(eng): Video: h264 (Baseline), yuvj420p(pc, bt709, progressive), 1920x1080, SAR 1:1 DAR 16:9, 60 tbr, 1k tbn (default)
  Stream #0:1(eng): Video: mjpeg, none(bt470bg/unknown/unknown), 1920x1080, SAR 1:1 DAR 16:9, 1k tbr, 1k tbn (default)

[akonanykhin]

@danleyb2 I don't think 60 is the correct fps for this video. VLC plays it at around 15, maybe even less, which feels correct. Is there no other way to determine fps?

[akonanykhin]

@danleyb2 Here is another video, which actually has FPS of ~17, but ffmpegcv reads it as 60 again:
https://drive.google.com/file/d/1PTVJ0QJj1mzzvMisfR-h-E78yQI8pxBp/view?usp=sharing

[akonanykhin]

@danleyb2 In this video ffmpegcv reads FPS as 25, when in reality it is 12:
https://drive.google.com/file/d/1RahC8Qr9tv4D4vUQ6-cpVSjyeQbS3nMR/view?usp=sharing

[danleyb2]

@akonanykhin I have been able to reproduce the issue, I am checking on an alternative way to get FPS.

[danleyb2]

@akonanykhin @marcbelmont Kindly review PR again
Fixed by calculating FPS manually by counting frames for 500ms