Great methodology! BUT,..what was the on set vfx sup thinking putting all those silly red markers in the first place? they do it all the time thinking its cool on set but causes more trouble that its worth :0
Год назад+1
can you help, I try to recreate your node setup carefully, everything seems fine, but after finish training, I click create inference then connect viewer to it, the screen is black and an error appears "Inference1: Error: Given model file doesn't exist or is not a.cat file"
Lemme grab my 2 RTX 8000 from the drawer real quick... Seems like Copycat needs a really powerful GPU to be time efficient. Otherwise i think doing it manually will be faster.. But it's a great start for machine learning in Nuke.
yeah, i mean this shot could be done with a gizmo like pixelfudger fill and it wouldn't take that long, but its cool to have this option, i think they will make a better optimization in the future
would prefer to have explanation of how you set up the nodes itself
Conheço Thiago Porto! Muito fera na Composição isso quando ele estava ainda no Brasil, foi ele que me inspirou a estudar vfx!
Great methodology! BUT,..what was the on set vfx sup thinking putting all those silly red markers in the first place? they do it all the time thinking its cool on set but causes more trouble that its worth :0
can you help, I try to recreate your node setup carefully, everything seems fine, but after finish training, I click create inference then connect viewer to it, the screen is black and an error appears "Inference1: Error: Given model file doesn't exist or is not a.cat file"
very informative. Thank you
alot of steps are missing. how do one understand how to setup the nodes
The corners of the photos look easier to track than the tracking markers.
Haha I agree. Only benefit I see is having markers in the middle to keep track of deformation
VFX artist removes the corners.
that shot look terrible difficult to do :O
Lemme grab my 2 RTX 8000 from the drawer real quick... Seems like Copycat needs a really powerful GPU to be time efficient. Otherwise i think doing it manually will be faster.. But it's a great start for machine learning in Nuke.
yeah, i mean this shot could be done with a gizmo like pixelfudger fill and it wouldn't take that long, but its cool to have this option, i think they will make a better optimization in the future
@@apmanti12 Agreed.
Very nice. Thank you mate. Good break down. Traditional methods would take a lot longer on the box,
a year later, a single 3090ti out performs 2 rtx8000 on my workstation in comparison, minus the Vram of course...
2x quadro rtx 8000 стоит как две моих почки. Очень "доступная технология"...