If you're training with a mask or using Warp to Landmarks, then you need an alignments file, as stated here:
Want to know about the Faceswap's Face Extraction process? Got tips, ideas or just want to learn about how it all works? Then this is the place for you
This forum is for discussing tips and understanding the process involved for Extracting and preparing face sets for training a model in Faceswap.
If you have found a bug are having issues with the Extraction process not working, then you should post in the Extract Support forum.
Please mark any answers that fixed your problems so others can find the solutions.
Great, that helps. I'm just getting into learning the Manual tool now, and made a decision to figure out masks and landmarks after I get the basics of the Manual tool, so I haven't actually looked into them yet. I'll postpone further comments on this question until I do.
Thanks to facewap for providing such a magical application, when I learned to extract according to the guide, I found that there are some abnormal situations, I don't know if other friends have encountered:
Extract the face in the video, the face in several frames of the picture is rotated at an angle, which will cause the face position to be misaligned in the subsequent training;
Among the extracted results, there are still pictures that are not faces. Can I directly delete them manually, and then use the “Clean the Alignments File” in the guide to process?
I would like to consult how to solve this situation.
Thanks ~ ···
The slightly misaligned faces can be fixed with the manual alignments tool.
My word is final
deephomage wrote: ↑Thu Jan 23, 2020 6:47 pm
To train a model, you need to extract both pairs of faces from video clips, to make a face set. For example, to swap Keanu Reeves and Nic Cage, you would extract Keanu and Nic faces from a variety of sources, to make a training set for each. If the video clips also contain unwanted faces, just delete the unwanted faces.
HI, regarding this, I have a few source videos of myself.
As you mentioned, I should extract my faces from these source; but do I output them to the same folder?
Code: Select all
And same for whoever I want to swap my face with?
Code: Select all
Just verifying and trying to understand whether I am on the right path.
- Site Admin
- Posts: 695
- Joined: Fri Jul 12, 2019 12:49 am
- Location: San Francisco
- Has thanked: 3 times
- Been thanked: 169 times
[mention]meDaniel[/mention] Yes, you can do that. You'll just need to follow the merging instructions in the guide to merge your alignments files after you finish extracting. Personally, I extract to different folders, then merge, but that's just a personal choice.
You definitely should be outputting to different folders, and then looking to fix up the alignments for each extract run prior to merging.
Once the files are merged, you won't be able to use the Manual tool to fix up any issues.
My word is final
Thanks for both the tips.
I will redo my extraction again. Because I extracted my own face from two source into the same folder.
Will try the method with the least issue =D
Guess I will be staring in some movies soon.
- Posts: 56
- Joined: Mon Sep 21, 2020 6:49 am
- Location: Italy
- Has thanked: 10 times
- Been thanked: 2 times
Hi, I'm trying the "Manually Fixing" function but I'm having trouble understanding what is right and what is wrong. Can you post photos? It would be easier. Thank you
I'm failing to understand exactly what I should have/keep from the extraction process to be used in either training or conversion. For example, following the extraction guide, for my "B" face(the face to be inserted INTO the video), when I finished, I now have two folders.
"B_faces_source" - which contains the original unaltered images to get faces from, as well as the alignments file I ended up with after following the guide.
"B_faces_extracted" - which contains the cropped/rotated face images generated during the extraction process and NO alignments files.
Once I get to the training section, it specifies that I should be feeding in the face files from the extracted folder, and not the source folder. But what about the alignment file? Can I use the alignment file from the source folder with the images from the extracted folder? Or do I feed the trainer both the original uncropped images from the source folder and that alignment file, and not actually use those cropped images in the extracted folder at all? When I did my extracting I used a mask (vgg-clean), so I know I definitely need to train with an alignments file.
First, again a BIG thank you for all the Faceswap team's hard work designing and supporting this massive project. We should all be supporting the team by signing up to Patreon.
Second, can I ask a question about alignments for conversion please?
I have an alignment file for training where I painstakingly adjusted each alignment mapped to the face. This extract used an EEN of every 10 frames or so.
I have now extracted a full alignment file for conversion with EEN 1. Is there any way to import the training alignments which I worked so hard on? I could then use those corrected alignments as a basis to correct the other 90% of "conversion alignments"?
Thanks very much for advice
bryanlyon wrote: ↑Mon Dec 16, 2019 3:49 pm
No, this is really not how the alignments are used. Even when copying from a previous frame, we recommend just using that as the starting point of detecting so that you get valid alignments for the face, otherwise you are just adding bad data that will confuse the model.
There are times when the mask is not on the face because the person is moving and wasnt properly detected. It would be helpful if moving the initial mask points in the manual frame would flow and carry over to the rest of the thumbs in manual, or a way to just have one mask in a frame say be copied to a specific number of other frames. Im talking about those points we can move around.