That’s actually possible by checking “apply modifier” in the collada export options:
http://wiki.blender.org/index.php/Dev:Ref/Release_Notes/2.64/Collada
Cool I will look into that. But it seems there were lots of changes, so I don’t know If I will have time to port all the changes into Mitsuba 4.0
Wenzel Jakob… WHAT. A. STUD.
I’ve been hacking a bit on the exporter (see my previous post) just to see if I could get something going. After spending a day trying to get OpenCollada to build and compile with Blender* there’s an issue I don’t know what to do about yet. The exporter calls a binary called ‘mtsimport’ that don’t exist anymore. Also I’m not certain the F17 Mitsuba rpm is build with support for Collada format, which is needed for this plugin to work. Again I just do this for kicks, if anybody want to run with it please do!
- If anybody is interested, the old Blender wiki page on compiling OpenCollada+Blender is outdated. First up use the most up to date revision of OpenCollada from here:
https://github.com/KhronosGroup/OpenCOLLADA
Building with Scons doesn’t work, only CMake, and the Expat library is not needed anymore (defaulting to libxml2).
Blender also must be build with CMake.
Looking at the release logs, it looks like its use of the latest in research has gotten it to the point where it can render just about anything, and at a good speed too.
Too bad the Blender foundation, doesn’t have the money to hire Wenzel the moment he graduates from college because these kind of algorithms would almost give Cycles the equivalent of the holy grail in pathtracing capabilities.
And for those still wondering, the reason he’s doing his own project as opposed to helping develop an established one is the fact that this is his research renderer, who knows what will happen to it when he graduates (especially if he gets snapped up by a company like Chaos Group, Random Control, and Nextlimit)
Since he’s a PhD student at Cornell (one of the leading universities in computer graphics) I assume he’s pursuing an academic career and want to continue to research. This enable him to continue to work on mitsuba, which would be great for other open source render engines. Doubt he would be interested to work on cycles unless he’s offered a ridiculous sum of money…
Has anyone had any luck installing this on Linux (Ubuntu 64 bit)
I downloaded the .debs from the Mitsuba links but they won’t install. Well actually the Collada files will but no the Mitsuba files. The Mitsuba dev indicates “dependencies not satisfiable” and the other will simply not connect.
I’d love to take this for a spin:)
Good thing is that WJ is doing his research in the open and letting people learn from his findings.
Brecht, Storm_st, Agus, etc. have good material if they at some point “find the need” to implement new things in cycles 
Now that i’m on a new distro that helps me keep my software current, I can finally play with Mitsuba for the first time, and I love it. There’s so much to discover, and the documentation is outstanding. Thanks Wenzel for making your research and work available for everybody!
I saw it has Python bindings, also for scene construction, so it might be possible to get scene data from Blender to Mitsuba directly without file ex/import. I haven’t looked at the most recent Blender plugin yet, maybe it works that way already, I still have to look where to find it. 
I’m playing with Mitsuba on win7 64 bit, works fine. How can I get access to the command line to change shaders etc? Anyone could help?
EDIT: Nevermind I solved i guess
How can I use my blender materials in here? I’m using the Blender exporter but it always renders all white.
You can’t, it uses its own distinct shaders that can only be edited directly in the xml scene files until someone creates an exporter.
Testing the bi-dir integrator (256 samples), i changed the default shader with a dielectric and the environment map with a studio hdr of mine. Seems really powerful, i wish a GUI so much.
bi-dir work only when light source can generate rays, as any light source except sun and background. To see bidir in full power you better use mesh light or point. With background as only light source you get pure unidirectional path tracer despite engine used (it only add some small CPU overhead).
storm you misunderstood what i meant, i couldn’t even talk with you about bi-dir or uni-dir, since i don’t know that stuff.
I was referring to Mitsuba power, seems really powerfull (Mitsuba). Thanks for the advice, i noted it does not produce caustics indeed, but i’m not able right now to get a lamp inside that scene.
yeah, but the mitsuba exporter isn’t automating that. i’d prefer not to have to write an xml file by hand for a scene after exporting a respective .dae file just to check “apply modifiers” 
on a different note, as other people have noted, the newest version isn’t playing nice with the old exporter. there are just so many changes. i thought about trying to fork the exporter for the newest version, but with my still limited knowledge of Python i’d probably end up making a fairly broken Frankenstein’s monster of an exporter. aren’t there any solid coders with a little time on their hands?? 
if anyone gets a working exporter for windows let me know… (even if just a simple one)
wouldn’t mind trying this out.
its is a neat render engine, would nice if the project had more resources as it seems promising.
Mind reader.
Mind you i find windows slower.
I will have exporter working this weekend. I will let you know when it is done.
That’s awesome. Can’t wait!