better result
Posted: Fri Mar 24, 2023 7:53 am
I'm looking forward for better result
Now I'm using DFL-H128 model. I cannot switch to another model due to not enough VRAM. (GTX 1050 Ti 2GB)
output extract face: 1024 (Both)
After Training awhile the overall result is 6.5/10.
I Already use no warp option to get better detail.
All face landmark is good but it look blur.
If I training more it will improve result ? training more with no warp ? how many Iteration ?
Or add more A/B Face ?Or any suggestion ?
if face alignment is inaccurate (When open wide mouth It's align only top lip).
- the result will come out with inaccurate same with alignment right ?
- what will happen to result if on/off warp to land mark option if inaccurate alignment ?
- I have many alignment face to fix, I have tried manual alignment but It such a pain there are a lot. Any suggestion to fix or make it faster and easier ?