AutoFrame feature for software about python-tf-bodypix HOT 4 CLOSED

Hi @benbatya thank you for your suggestion.
This seems to have an interesting feature to have.

For now I would suggest for a feature like this to live outside this repository.
I would like to keep this project relatively light to encourage others to include it, if they only want to use the bodypix model.

I had started layered-vision, which has a broader scope. So perhaps that could be a good home?

I should also mention virtual_webcam_background (GPL licensed), which is a project I used, before trying out my own implementation. It doesn't currently use tf-bodypix, but there is an issue #47 for that.

What do you think about starting with a separate script (gist?) to experiment with your suggested feature and then we could see how to best integrate it? I guess it might also need a bit of #50 (although just looking for the face should be easier)

from python-tf-bodypix.

Hi @de-code sure, here's a gist. I hope that the patch format works.
https://gist.github.com/benbatya/59e717c50a22e1e8278918d04bb06cd6

Yeah, I'll check out layered-vision vs virtual_webcam_background.
I am curious if you started layered-vision because of issues with VWB or more to scratch a coding itch.
I'm happy to implement the feature in either project, it's a fairly simple additional filter...

from python-tf-bodypix.

After looking at layered-vision, it's a good framework. I'll implement the filter there.
Tracking issue is de-code/layered-vision#56

from python-tf-bodypix.

I am curious if you started layered-vision because of issues with VWB or more to scratch a coding itch.

Yes, it was a combination of things probably. In part I wanted to try a quite different architecture. I thought a kind of graph like architecture may allow for more optimisation later on (e.g. only calculate layers or parts needed).

from python-tf-bodypix.