This forum is for reporting errors with the Convert process. If you want to get tips, or better understand the Convert process, then you should look in the Convert Discussion forum.
Please mark any answers that fixed your problems so others can find the solutions.
06/16/2022 16:58:29 VERBOSE Found more than one face in an image! 'test_007039.png'
06/16/2022 16:58:29 VERBOSE Found more than one face in an image! 'test_007043.png'
06/16/2022 16:58:29 VERBOSE Found more than one face in an image! 'test_007044.png'
06/16/2022 16:58:29 VERBOSE Found more than one face in an image! 'test_007045.png'
06/16/2022 16:58:30 VERBOSE Found more than one face in an image! 'test_007047.png'
06/16/2022 16:58:30 VERBOSE Found more than one face in an image! 'test_007048.png'
06/16/2022 16:58:30 VERBOSE Found more than one face in an image! 'test_007050.png'
06/16/2022 16:58:40 INFO -------------------------
06/16/2022 16:58:40 INFO Images found: 7050
06/16/2022 16:58:40 INFO Faces detected: 14100
06/16/2022 16:58:40 INFO -------------------------
06/16/2022 16:58:40 INFO Note:
06/16/2022 16:58:40 INFO Multiple faces were detected in one or more pictures.
06/16/2022 16:58:40 INFO Double check your results.
06/16/2022 16:58:40 INFO -------------------------
06/16/2022 16:58:40 INFO Process Succesfully Completed. Shutting Down...
Prior to the latest update the convert tool would swap all faces in a frame as long was they were in the alignments file. After the update the convert tool only swaps one of the faces even when all faces are shown in the manual tool. I know that you normally only want a single face swapped but occasionally there is a reflected face in a mirror or a photo of the same person in the room that you also want to swap. Is the change in behaviour intentional?
I can confirm this effect of swapping only one face per frame/picture if multiple faces are in the alignment file.
I also found the root for this effect: lib/convert.py in line 248 the following change was successful:
"borderMode=cv2.BORDER_CONSTANT" -> "borderMode=cv2.BORDER_TRANSPARENT"
Hope I could help to fix this in the next update. Until then, the fix can be done locally in the faceswap folder.
I can confirm this effect of swapping only one face per frame/picture if multiple faces are in the alignment file.
I also found the root for this effect: lib/convert.py in line 248 the following change was successful:
"borderMode=cv2.BORDER_CONSTANT" -> "borderMode=cv2.BORDER_TRANSPARENT"
Hope I could help to fix this in the next update. Until then, the fix can be done locally in the faceswap folder.
Good Luck!
Thanks for this, this is useful information. It's also annoying as changing the border mode fixed a bug where a distinct line could be seen at the edge of the face box. I will look into this to see if I can find a better solution.
The actual program you could have during the initial reading of face folder to use during swap.
You could have a multi-face found setting which appears -> drop down
Swap All / Swap 0 / Swap 1 -
Of course having both [ 0, 1 ] in there in first place would also be a valid check which apparently was doing that in past. I am like other person are replacing 3D movie frames ( LEFT / RIGHT) images so I always choose All. But I could see someone trying to have two people next to each other talking have two faces per frame want to target one or other but probably limited use compared to replacing ALL.
Yeah, I understand that if the algorithm examining frames doesn't start from x left to right but simply finding the face whichever one it finds first outputting first found as _0 and next _1 next _2 so forth. Hope the double face issue can be solved though to work around either I have to revert back or right now I am trying on current update, Just to rerun the MOVIE output two times which I haven't confirmed yet if it works. But the when one frame finds first Face it's getting replaced then I taking that movie rerunning it now hoping to replace it second run. Maybe it was my alignment file but I tried to run it again didn't work second time. So I am re-extracting make sure to reuse old alignment file and [x] skip existing faces [x] skip saving faces. (unless skip existing faces means skipping fames to which I assume skip existing was) See if works. If not. I guess I have to find old branch of program in meantime. Unless it's possible to flip branch version inside here looks like you have a few branch versions built inside the program.