Re: [Gimp-developer] GSoC 2013 - n-point image deformation

[Marek Dvoroznak <dvoromar gmail com>]

Hello Sven,

thanks for the reply.

> How does the proposed method work
> - with more complex objects like a linear rope or cartoon drawings?
> (See for instance some puppet warp examples on [1])

I have looked at examples.

The proposed method is very good for bending objects or manipulating
articulated objects - doing deformation of that tree in examples
shouldn't be a problem for this method.

It's not so good in shrinking of objects - but there can be an easy
extension added to it which can handle shrinking. It has some
limitations, though. I think that Bounded Biharmonic Weights method is
better at that but it has some other disadvantages.

Sometimes it would be useful if the deformation behaved similar to
Moving Least Squares method (Fiji, Krita). Again with an easy extension
(using weights from MLS) that is not a problem - it behaves very
similarily.

I'm attaching a video clip [1] presenting these extensions on image of
Mona Lisa.

> - work with non-solid and non-transparent backgrounds like in photographs?
> Will it fill the uncovered parts of the image in a content-aware manner 
> or can it be integrated with - for instance -
> the Resynthesizer plugin [2] to achieve this?
> Will the result still be plausible regarding lighting or can the user do 
> this easily afterwards?

If we want to move subject's parts (e.g. arms as in Sintel example [2])
freely we always have to cut it or select it. It's the same as in
Photoshop - to remove a background, something like Content Aware Fill in
Photoshop or Resynthesizer in GIMP has to be used.

> - in comparison to GIMPs cage transform?

Cage transform should be more precise (when every pixel is manipulated)
but it's suitable for different kinds of deformations.

> It would also be good if the tool worked with high bit color depth and 
> thus were in GEGL
> and already connected to the GIMP UI.

It should use GEGL but I don't know how fast it will be - I don't have
so much experience with it. But I assume that every additional layer has
it's overhead. Do you have any experience with that?

I wanted to make a plug-in first but I feel like UI in plug-in is not so
user-friendly. I don't know...

Regards,
Marek

[1] http://youtu.be/0_nl9fSTfMY
[2] http://youtu.be/ieaHvHAcNRE