Faceswap Forum

Hi there,

We are still playing around with some VR projects at the moment.

We're noticing something with regards to extracting faces/alignments.

Some of the footage we extract for both training and conversion is 3D side-by-side VR footage. Essentially, two videos side by side - one for the left eye and one for the right eye.

When extracting, and when the faces are quite far away from the capturing lens, the faces look almost identical and it seems that the extraction process thinks they are the same face and only extracts one of them.

This is fine for training sets but when it comes to the final conversion, we need both faces swapped out (on both sides of the video).

Are there any tweaks we can use to force the extract process to capture all faces even if they look identical?

Thinking about it - in theory, we could split the video into left/right eyes and process them individually... this is an option of course but it would increase workload.

Thank you for your help in advance.

The extraction process makes no comparison on faces. It will extract all faces it finds in a video. If it's not extracting one of the faces then it's not finding the face.

Unfortunately I have no experience faceswapping with VR, so this is about the extent that I can help, I'm afraid.

Thank you for that reply [mention]torzdf[/mention] and confirming the extraction process doesn't only extract the one (same) face.

That's also interesting information and puzzles me as to why only one of the faces is extracted - because both sides of the video - left/right clearly have a very visible face on both of them.

We are going to try splitting the video down the middle and running two extraction processes (left/right) to see if we get different results.

Will update this post with results for anyone interested.

Just an update:

Splitting the video into two separate videos has fixed the issue - now extract picks up the "missing" faces. Perhaps it was also something to do with the frame size of the input video (5K).