Unavailable in Zoom login screen, altough works in the real meeting

[raffaem]

It works perfectly if I see the webcam with

mplayer tv://device=/dev/video0

But it doesn't work with Zoom web interface, it says "Video unavailable at this time"

I have Firefox 97 on Fedora Workstation 35.

I use this to start the script:

$ python3 virtual_webcam.py 
2022-02-17 08:57:04.326164: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory
2022-02-17 08:57:04.326184: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
Reloading config.
2022-02-17 08:57:05.550383: E tensorflow/stream_executor/cuda/cuda_driver.cc:271] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected
2022-02-17 08:57:05.550418: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:169] retrieving CUDA diagnostic information for host: fedora
2022-02-17 08:57:05.550426: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:176] hostname: fedora
2022-02-17 08:57:05.550472: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:200] libcuda reported version is: 510.47.3
2022-02-17 08:57:05.550494: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:204] kernel reported version is: 510.47.3
2022-02-17 08:57:05.550500: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:310] kernel version seems to match DSO: 510.47.3
Model: mobilenet (multiplier=0.5, stride=16)
Loading model...
2022-02-17 08:57:05.690041: I tensorflow/core/platform/cpu_feature_guard.cc:151] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2022-02-17 08:57:05.710284: W tensorflow/core/common_runtime/graph_constructor.cc:1511] Importing a graph with a lower producer version 38 into an existing graph with producer version 987. Shape inference will have run different parts of the graph with different producer versions.
2022-02-17 08:57:05.737381: I tensorflow/core/grappler/devices.cc:66] Number of eligible GPUs (core count >= 8, compute capability >= 0.0): 0
2022-02-17 08:57:05.737598: I tensorflow/core/grappler/clusters/single_machine.cc:358] Starting new session
2022-02-17 08:57:05.770818: I tensorflow/core/grappler/optimizers/meta_optimizer.cc:1164] Optimization results for grappler item: graph_to_optimize
  debug_stripper: debug_stripper did nothing. time = 0.025ms.
  remapper: Graph size after: 244 nodes (0), 171 edges (0), time = 0.602ms.
  constant_folding: Graph size after: 172 nodes (-72), 171 edges (0), time = 3.895ms.
  arithmetic_optimizer: Graph size after: 172 nodes (0), 171 edges (0), time = 1.635ms.
  dependency_optimizer: Graph size after: 172 nodes (0), 171 edges (0), time = 0.744ms.
  debug_stripper: debug_stripper did nothing. time = 0.013ms.
  remapper: Graph size after: 172 nodes (0), 171 edges (0), time = 0.388ms.
  constant_folding: Graph size after: 172 nodes (0), 171 edges (0), time = 2.121ms.
  arithmetic_optimizer: Graph size after: 172 nodes (0), 171 edges (0), time = 1.728ms.
  dependency_optimizer: Graph size after: 172 nodes (0), 171 edges (0), time = 0.787ms.

done.
/run/media/raffaele/55ab61c4-83cf-4d9f-a5cd-7fcfdc14b4fb/data/progetti_altrui/virtual_webcam_background/virtual_webcam.py:187: DeprecationWarning: `np.float` is a deprecated alias for the builtin `float`. To silence this warning, use `float` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.float64` here.
Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations
  frame = frame.astype(np.float)

Here is my config.yaml file:

$ cat config.yaml
segmentation_threshold: 0.25
blur: 0
erode: 0
dilate: 0
virtual_video_device: "/dev/video0"
real_video_device: "/dev/video1"
average_masks: 3
mjpeg: False
layers:
  - "empty": [["image", "/home/raffaele/Pictures/webcam_background.png"]]
  - "foreground": []

[allo-]

Do you have the problem only on the login screen? I can confirm that the login screen looks like this, but after login it works just fine.

[raffaem]

Ah I don't know, the meeting has not started yet

[allo-]

At least for me it is like that. I do not know why. I think the preview finds the wrong camera and tries the (blocked) real device, but I have no idea why it works different from the real meeting. But as long as it works in the meeting I think it's no big deal.

Any advice how to fix it for the preview is welcome, of course.

[raffaem]

Yeah it works in the real meeting

[allo-]

Would you like to help testing?

You could try if the same happens with other streams and v4l2loopback: https://atinkerholic.wordpress.com/2018/10/10/how-to-use-a-virtual-webcam-with-static-image-or-video-ffmpeg-v4l-utils-and-v4l2loopback/

You could also test with akvcam instead of v4l2loopback. One of the main differences is, that akvcam generates frames when there is no input, while the loopback device is more a FIFO buffer for frames, which keeps the browser waiting for frames when the virtual webcam is not sending frames fast enough.

I think chromium had some threshold like 5 FPS under which it doesn't recognize a cam anymore, but I suppose that that's not your problem. At least I see a similar issue and on my PC the frame rate is high enough.