Frame Loss in video stream

[Rollson]

❔Question

Hi, thank you for your great work:). Please I have 2 questions.

1. detect.py stops running any time there is a frame loss in the HTTP video stream. I tried modifying the datasets.py to be able to handle frame loss but I couldn't fix it. Please what part of the datasets.py or detect.py can I modify to fix this frame loss issue and any ideas on how to do it?

2. Also which is the proper way to specify the --source argument in detect.py. Is it
   python detect.py --source "http://"
   or
   python detect.py --source http://

why am asking is because when the URL is not specified as a string (i.e when I run it this way: python detect.py --source http:// ) , detect.py only run for some seconds and then stops.
Thanks for your response:)

Additional context

[glenn-jocher]

@Rollson thanks for the notice. Please supply a Colab notebook with a reproducible example that we can look at.

String encapsulation is up to you and has no effect on inference.

[Rollson]

@glenn-jocher. thanks for your response. anytime there is a frame loss in the stream, it shows this error:

img = [letterbox(x, new_shape=self.img_size, auto=self.rect)[0] for x in img0]
File "/home//Dokuments/yolov5/utils/datasets.py", line 696, in letterbox
shape = img.shape[:2] # current shape [height, width]
AttributeError: 'NoneType' object has no attribute 'shape'

[glenn-jocher]

@Rollson so it seems that if there is a dropped frame, img will be equal to None, and this error will occur. Is there a way to reproduce this dropped stream event so we can test a fix?

[Rollson]

@glenn-jocher there is no way I can reproduce the dropped stream apart from testing with the stream url

[glenn-jocher]

@Rollson ok understood. If you can debug and produce a fix that's working on your side then, we'd be happy to review it as a PR.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[glenn-jocher]

@Rollson this same issue identified in #2155 and #2196. We should have a fix soon, though quite late for you, apologies.