Output is really weird and unexpected

[filipe-monteiro]

Lovely project but when trying to run it the output was unexpected.

I got some user warnings but similar to the ones in the jupyter notebook shared in the repo.

All the outputs look like this:

Any idea of what I did wrong?

[ZZZColin]

Congret, you just made another TO-BE-WORLD-FAMOUS painting

[anandpawara]

Lovely project but when trying to run it the output was unexpected.

I got some user warnings but similar to the ones in the jupyter notebook shared in the repo.

All the outputs look like this:

Any idea of what I did wrong?

You did nothing wrong the output of the image depends on first frame frame opencv, If the user is not correctly oriented or not present, then the because of first frame , the application outputs bizzare results need to work on taking best frame of refrence before generating output

[Otter-man]

@anandpawara solution for this problem might be easier. Maybe add the "reset" button, that will make the reference frame the current one. So if the user is not correctly oriented or not present he can reset the reference frame on the fly.

Will be like this: User seeing distorted image will position himself correctly, will press there "reset button" and the generation will restart using current frame (at the time of the button push) as a ref.frame.

This solution is used in other real-time deepfake generator project: https://github.com/alew3/faceit_live3

I'm trying to implement this myself but i'm not that familiar with opencv and can't yet figure how to recreate it in your code. Maybe you'll be able to implement it yourself faster.

[anandpawara]

@anandpawara solution for this problem might be easier. Maybe add the "reset" button, that will make the reference frame the current one. So if the user is not correctly oriented or not present he can reset the reference frame on the fly.

Will be like this: User seeing distorted image will position himself correctly, will press there "reset button" and the generation will restart using current frame (at the time of the button push) as a ref.frame.

This solution is used in other real-time deepfake generator project: https://github.com/alew3/faceit_live3

I'm trying to implement this myself but i'm not that familiar with opencv and can't yet figure how to recreate it in your code. Maybe you'll be able to implement it yourself faster.

This is exactly I was planning to make in future , thanks for the lead may I'll try to integrate with current project

[filipe-monteiro]

Thanks everyone, I see what I need to do now!

Will close this issue.