mask accuracy in relation to higher face coverage e.g. 87% legacy

Want to understand the training process better? Got tips for which model to use and when? This is the place for you


Forum rules

Read the FAQs and search the forum before posting a new topic.

This forum is for discussing tips and understanding the process involved with Training a Faceswap model.

If you have found a bug are having issues with the Training process not working, then you should post in the Training Support forum.

Please mark any answers that fixed your problems so others can find the solutions.

Post Reply
User avatar
swapration
Posts: 21
Joined: Thu Sep 10, 2020 1:21 am

mask accuracy in relation to higher face coverage e.g. 87% legacy

Post by swapration »

1. Is it bad for masks to cover more than the edges of a face (backgrounds, hair, .etc) for models with higher coverage - for capturing a more accurate silhouette e.g. the bridge of the nose, cheekbone protrusion, etc.? Will it just add to the training time or will it draw things outside of the face during conversion?

2. Do higher coverage models learn beyond what the mask includes? For example, with unet masks ending just above the brow, will a coverage of 87.5% (legacy) still learn some of the forehead?

3. Also is the coverage the same between the new faceswap (extracted faces end near the collar) and the older version (extracted faces end below the chin)? Is 87.5% still down to the chin for the new version of faceswap like it was for the older version?


User avatar
torzdf
Posts: 1438
Joined: Fri Jul 12, 2019 12:53 am
Answers: 125
Has thanked: 47 times
Been thanked: 278 times

Re: mask accuracy in relation to higher face coverage e.g. 87% legacy

Post by torzdf »

swapration wrote: Sat Jun 05, 2021 10:45 pm
  1. Is it bad for masks to cover more than the edges of a face (backgrounds, hair, .etc) for models with higher coverage - for capturing a more accurate silhouette e.g. the bridge of the nose, cheekbone protrusion, etc.? Will it just add to the training time or will it draw things outside of the face during conversion?

For training, I don't worry too much (I figure enough masks will be correct that it will average out over time).

For convert the mask needs to be correct, bad masks will generally lead to bad converts

swapration wrote: Sat Jun 05, 2021 10:45 pm
  1. Do higher coverage models learn beyond what the mask includes? For example, with unet masks ending just above the brow, will a coverage of 87.5% (legacy) still learn some of the forehead?

No. Only what is in the mask is learned. If you have a large coverage, but the mask only includes a small part in the middle of the image, then this is all that will be learned.

swapration wrote: Sat Jun 05, 2021 10:45 pm
  1. Also is the coverage the same between the new faceswap (extracted faces end near the collar) and the older version (extracted faces end below the chin)? Is 87.5% still down to the chin for the new version of faceswap like it was for the older version?

It is for 'legacy' centering. For different centering different coverage will be required. See here: viewtopic.php?f=27&t=146#config_global

My word is final


User avatar
swapration
Posts: 21
Joined: Thu Sep 10, 2020 1:21 am

Re: mask accuracy in relation to higher face coverage e.g. 87% legacy

Post by swapration »

I'm guessing by 'convert the mask needs to be correct' do you mean for face A, B, or both? I'm guessing you mean face A needs to be accurate due to obstructions?


User avatar
torzdf
Posts: 1438
Joined: Fri Jul 12, 2019 12:53 am
Answers: 125
Has thanked: 47 times
Been thanked: 278 times

Re: mask accuracy in relation to higher face coverage e.g. 87% legacy

Post by torzdf »

I mean the masks stored in the alignments file that relate to the video you will be converting on to.

My word is final


Post Reply