questions about the deformation degree about shapematchinggan HOT 7 CLOSED

The ``dynamic'' here does not means changing the deformation degree, it means that the water effects change within the same deformation degree.
(1) To make it more deformed, just set ''--scale 1'', the maximam deformation degree in our model.
(2) I just linearly map the standard deviation of gaussian blur ranging from 8 to 24 into a normalized range of [0,1], which is easy to understand and adjust. You can map it to any ranges as you like.
For more details, please refer to the paper.

from shapematchinggan.

Thank you very much. But I don't understand this " ”dynamic'' means that the water effects change within the same deformation degree".
I think changing the degree of deformation and concat the corresponding generated images are to create a dynamic feeling(like .gif file).
Maybe I am wrong. So how did you generate this file ./imgs/teaser-c.gif？

from shapematchinggan.

Please see the Section 5.1 Network architecture.
''Since the structure map contains many saturated areas, we add gaussian noises onto the input of GS
and GT to avoid ambiguous problem. It also empowers our network to generate diversified results during testing.''

I just add random noise to the text images so that each time I will get a different results even I use the same deformation degree.
If you want to get a continously changing animation, you can just use two random noises, and interpolate them to get a set of noises to generate each frame of your animation.

from shapematchinggan.

So the reason why the animation effect can be generated is not because of changing the degree of deformation, but because of the addition of random noise?

from shapematchinggan.

Yes. I think you might be confused with dynamic typography and controllable text stylization.

• The ''dynamic typography'' (teaser-d.gif, teaser-e.gif, teaser-f.gif) is an application of our method, by setting different noises but the same l.

• Our main contribution is the ''controllable text stylization'' (teaser-c.gif), by setting different l but the same noise.

from shapematchinggan.

Yes, I was a bit confused before, and now I understand. Thank you very much. ^_^

from shapematchinggan.

You are welcome.
Thank you for your interest in our work :)

from shapematchinggan.