I try to run demo_webcam.py on cpu, linux. But I get some question, eg: OSError:[Error 22] invalid arguments FATAL: exception not rethrown Abolished (core saved) I'm not sure what went wrong. Support only NVIDIA？ Looking forward to your reply！！！ Thank you！

This project is really great. Great job all-around.

Problem:
Understanding how the different model checkpoints and parameters impact the quality of the model is difficult to discern. These must be specified ahead of time, making it difficult to test various combinations of parameters and checkpoints in an easy fashion. One would have to start and stop the video for each change, which can make it difficult to compare quality levels.

Solution: Enable the model parameters and checkpoints to be modified in real time so one can see the immediate impact of the changes.

To accomplish the above, I expanded the control panel window to enable the appropriate buttons for tweaking the model. This required a bit of refactoring, but I tried to preserve the original simplicity of the code. I also did some light reformatting, fixed an issue where playing videos would crash, and hid all buttons until a user selects a background. Example screenshot of the updated control panel below.

The UI changes depending on the refine mode selected. The "Full" refine mode, for example, hides the row of buttons beneath it since those wouldn't apply. To enable switching of the model checkpoint I added a new optional --model-checkpoint-dir parameter where a user can specify the directory which contains the model checkpoint(s) that have been downloaded. The code will look for known models based on name in that directory and enable the user to cycle through them. If the --model-checkpoint-dir is not specified, then the bottom button is hidden.

This request contains some minor changes to add Windows support for BGMv2 plugin using pyvirtualcam module.

demo_webcam_win.py and installation description added.

I have installed newer version of pytorch, than asked, as it was no longer available. I have 2 fake devices (video0-1) and will be sharing logs of attempts on both of them. Sudo or not, output is the same.

Command 1: sudo python demo_webcam.py --camera-device /dev/video0 --model-checkpoint torchscript_resnet50_fp16.pth

Output: [ WARN:0] global /tmp/pip-req-build-ddpkm6fn/opencv/modules/videoio/src/cap_v4l.cpp (1004) tryIoctl VIDEOIO(V4L2:/dev/video0): select() timeout. Traceback (most recent call last): File "/home/albert/git/BGMv2-webcam-plugin-linux/demo_webcam.py", line 586, in fake_camera = pyfakewebcam.FakeWebcam(args.camera_device, cam.width, cam.height) File "/usr/lib/python3.9/site-packages/pyfakewebcam/pyfakewebcam.py", line 54, in init fcntl.ioctl(self._video_device, _v4l2.VIDIOC_S_FMT, self._settings) OSError: [Errno 22] Invalid argument

Command 2: sudo python demo_webcam.py --camera-device /dev/video1 --model-checkpoint torchscript_resnet50_fp16.pth

Output: [ WARN:0] global /tmp/pip-req-build-ddpkm6fn/opencv/modules/videoio/src/cap_v4l.cpp (1004) tryIoctl VIDEOIO(V4L2:/dev/video0): select() timeout. Traceback (most recent call last): File "/home/albert/git/BGMv2-webcam-plugin-linux/demo_webcam.py", line 594, in if app_step(): File "/home/albert/git/BGMv2-webcam-plugin-linux/demo_webcam.py", line 524, in app_step frame = cam.read() File "/home/albert/git/BGMv2-webcam-plugin-linux/demo_webcam.py", line 73, in read frame = self.frame.copy() AttributeError: 'NoneType' object has no attribute 'copy'

All I have done is update torch to 1.8.1 and set cudatoolkit to 11. I could not install torch using pip from the instructions page on pytorch.org for CUDA 11, which is required for Ampere, so I used Conda.

My Garuda Arch couldn't install the thing (some Nvidia issues), and Debian does this - and this is the issue -

python3 demo_webcam.py --model-checkpoint torchscript_mobilenetv2_fp16.pth
Traceback (most recent call last):
  File "/root/demo_webcam.py", line 586, in <module>
    fake_camera = pyfakewebcam.FakeWebcam(args.camera_device, cam.width, cam.height)
  File "/usr/local/lib/python3.9/dist-packages/pyfakewebcam/pyfakewebcam.py", line 54, in __init__
    fcntl.ioctl(self._video_device, _v4l2.VIDIOC_S_FMT, self._settings)
OSError: [Errno 22] Invalid argument
FATAL: exception not rethrown
Aborted

I tried pyfakewebcam from both apt & git