Defining new backends (akvirtualcamera)

[ratuspro]

I've been playing with pyvirtualcam for the past couple of weeks and I'm currently facing the challenge of handling two virtual cameras that are not supported by the built in backends.

Unfortunately, I'm not able to figure out how to define in python other virtual cameras.
I've looked at the built in ones and they are all in cpp and there are no examples on how to create my custom cameras.

Btw, I'm using AkVCam to create new virtual devices.

[letmaik]

I think the word "defining" is not right here. It would be a new implementation for a new backend. And that typically requires writing native code. The existing backends written in C++ all follow the same structure, but obviously every backend needs different code to interact with a virtual camera.

So, someone would have to write a backend that can talk to the akv virtual camera. I'm assuming you're referring to the macOS/Windows one from https://github.com/webcamoid/akvirtualcamera. How that code will look like and how easy it is will become clear at implementation time.

Any reason why you can't use the other available backends?

[ratuspro]

You may find my progress on the following fork to add a AkVCam Backend.

As suggested, I'm taking inspiration from the other backends.
Currently, the backend is able to find the devices in the registry as well as their preferences (height, width, format, etc...)

However, I'm experiencing some problems while trying to bind to the devices IO.
I'm following the examples provided on akvirtualcamera repo (here). They launch a new process to handle the new frames.

I'm having problems defining struct StreamProcess because, during the setup, some of its fields types are not found (e.g. 'SECURITY_ATTRIBUTES'). I'm pretty sure it has something to do with the libraries linked in extension modules. However, I'm not familiar with setuptools nor pybind11.

If you could provide some assistance I would greatly appreciate.

[letmaik]

According to the docs it should be enough to include windows.h. Not sure why you're getting errors there.

[ratuspro]

I also though the same. Well, after changing to miniconda it works.

As it stands, the AkVCam backend is capable of:

• identifying what virtual cameras are available on a windows machine
• start streaming using the AkVCam supported formats (still missing other formats available for other backends)

I've still not worked on test, docs, or examples.
I added you to my fork as a collaborator.
If you prefer I can do a PR with this contribution and we can iterate on this new backend there.

[letmaik]

Great! Yes, I think a PR makes sense, then it makes it easier to review and add comments.
A few notes already here:

• The registry has the install path of akvcam, which should be used to construct the full exe path.
• Instead of C buffers, let's use std::vector<uint8_t>, then memory gets deallocated automatically
• Can a frame be written to stdin in full or does it have to be line by line? The former would be simpler
• The external process may exit at any point which needs to be handled
• It's fine to not support all image formats

[ratuspro]

Thanks for the feedback.

This can be found here #88