- This topic has 0 replies, 1 voice, and was last updated 1 year, 9 months ago by
ebirch3.
ebirch3.-
Posts
-
ebirch3
So I just started trying to do my first deep fake. I just had a few questions that is kind of confusing me during the face extract process.
So I have extracted the faces from my source material and that all seems to be working fine and I have deleted any wrong images. However currently I am trying to extract the face from my destination video and a lot of the images are showing up on an angle with black around the edges of the image as if it’s tilted to one side. I am guessing I will have to do the manual face extraction potentially for this video as it seems to be struggling. The destination video is basically me panning the camera up from the bathroom mirror which then shows a reflection of my face which I want to tranform into trumps face. So I think the problem maybe with the first few seconds of the video as the AI tries to find my face. Maybe I should cut the first few seconds of the video out entirely so it’s not trying to map onto like half a face as the camera tilts upwards.
My question really has a few points to it actually. The first is what is the data_dst alligned_debug for. As that folder seems to not have any errors in it really. The images for that folder show this almost transparent material (I think it looks like the Xseg feature that surrounds the face). The folder that seems to have issues though is the data_dst alligned as that’s the one showing the weird angles. So can I just delete those images that are wrong from the aligned folder? (as there are more than half id say that aren’t wrong). Will it be easy then to allign the rest using the manual face extraction. Like do I need to remember what frames are fine and which ones aren’t? Or when I click on the manual face extract from dst will it just only show me the ones which had been deleted? I am just wondering which process would be fastest for me. As if I just delete all the aligned and do it all manually that may take a lot longer.
Sorry for the giant wall of text. Any help would be greatly appreciated!
ebirch3- You must be logged in to reply to this topic.