[Bf-committers] Freestyle NPR for Blender: Call for developers

[Stéphane Grabli]

Hi,

The condensed version: We are looking for one or (preferably) several
developpers to integrate Freestyle, a Non-Photorealistic Renderer, to
Blender (possibly through a Blender-endorsed Google SOC project).

And now for some details:

Freestyle is a renderer that generates line-drawings from 3D scene;
similarly to Renderman, style is described through programmable
shaders. It is the experimenting platform of a research project
conducted jointly by teams from INRIA in France (artis.imag.fr) and
MIT in the States (graphics.lcs.mit.edu).  Currently, it is available
from sourceforge under the GPL license as a standalone program. Images
as well as a more thorough description of the software can be found at
this address:

http://freestyle.sourceforge.net/GALLERY/dirindex.php

A video presenting the approach as well as some results can be viewed
at this address:

http://artis.imag.fr/Membres/Emmanuel.Turquin/files/video-divx-sub.avi

Since its public disclosure, many people have been adventurous and
curious enough to try Freestyle and have manifested a positive
interest in such a renderer - to our very surprise, given its rather
confidential release, and its pre-alpha state.  Soon it became obvious
that Freestyle, still a "researchy" software, had to evolve to meet
the demands of a growing user basis. Amongst these users, many belong
to the Blender community and suggested that Freestyle would make a
nice complement to Blender. In particular, many artists have expressed
their enthusiasm through threads of the blenderartists.org forums, and
are rooting for a seamless integration in a future release of Blender:

http://blenderartists.org/forum/showthread.php?t=89352&highlight=freestyle
http://blenderartists.org/forum/showthread.php?t=67587&highlight=freestyle
etc...

Although Freestyle is only a first step on the road to a complete
procedural NPR renderer (including handling of temporal coherence), it
seems it would still position Blender ahead of existing commercial NPR
packages.

Unfortunately, none of the main developpers of Freestyle (read: the
two of us, Stephane and Emmanuel :) ) can currently carry on this
development without help, because of time constraints as well as a
lack of familiarity with Blender's sources. This email is intended to
hopefully generate some interest around this project so that some
developers that have both time and motivation could help us keep
Freestyle alive as an extension to Blender.

Freestyle is developed in C++ and interfaces with Python for the style
language (using SWIG to generate the interface). The current version
also requires Qt 3.X for the GUI, as well as an additional Qt-related
library called QGLViewer - dependencies that we would have to get rid
of.  In the current workflow, everything starts with the loading of a
3DS model, on which various geometrical computations are applied,
resulting in a graph carrying loads of information (aka the
"ViewMap"). We believe that all this part could be largely rewritten,
using the current Blender scene as input, and taking advantage of
Blender's own code for all calculations.  Once the graph is computed,
shaders expressed in Python are applied to it that generate the final
image: this second stage also fits in nicely in Blender, since its
scripting language is already Python.

The few points above barely scratch the surface of the work that has
to be done for an integration to Blender; not to mention, in addition
to purely technical issues, some more theoretical/metaphysical ones
would inevitably arise throughout the development cycle. One of this
thread's aims is precisely to discuss them and answer any question you
may have.

Originally, we thought this work would be a good candidate for a
Google Summer of Code project, sponsored by Blender. In that case, we
would serve as "mentors" for the developper willing to tackle on the
task. Needless to say this developper would have to be extremely
proficient and have a good knowledge of the Blender sources if the
project has to be completed in less than 3 months.

But of course, if more people are motivated for joining the effort
(...and for free, as in free beer :) ), it can be done completely
independently of the SoC... the more the merrier. This would also free
us from the tight 3-months constraint.

Thank you, we are looking forward to reading your reactions!

Stephane Grabli & Emmanuel Turquin

P.S.: We also started a thread on blenderartists.org to gather
reactions/comments from the artists themselves about this potential
integration:

http://blenderartists.org/forum/showthread.php?t=89986

Stephane Grabli & Emmanuel Turquin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://projects.blender.org/pipermail/bf-committers/attachments/20070223/0b0c1a7f/attachment-0001.htm