Getting frustrated now, whatever I do the image is blury. I have found what I think is good reference footage from a 20 min long interview, in a well lit video. Then I replicated the lighting and angle of that video for the A model, I have around 15000 images to work from, have tried original, Dlight and DFL-H128 and have trained up to 150000 iterations The images in the training look great but the output video looks like mush. I have seen others on this forum getting amazing results at 120,000 iterations so I have no idea what else to do or what I am getting wrong.
blury image, no mater what I do.
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.
- darkfable12
- Posts: 6
- Joined: Wed Jan 06, 2021 1:54 am
- Has thanked: 2 times
Re: blury image, no mater what I do.
It nearly always comes down to Data. If there is anyway you can post a screengrab of the preview when training, we may be able to give you more insight.
Also "matching lighting" is a bit misleading. You want variety rather than consistency.
My word is final
- darkfable12
- Posts: 6
- Joined: Wed Jan 06, 2021 1:54 am
- Has thanked: 2 times
- darkfable12
- Posts: 6
- Joined: Wed Jan 06, 2021 1:54 am
- Has thanked: 2 times
- darkfable12
- Posts: 6
- Joined: Wed Jan 06, 2021 1:54 am
- Has thanked: 2 times
Re: blury image, no mater what I do.
this is at 50,000 iterations with DFL - H128 at batch size 65 it seems to be wavering close to convergence.
- darkfable12
- Posts: 6
- Joined: Wed Jan 06, 2021 1:54 am
- Has thanked: 2 times