[Bf-committers] Importing Assets: FBX VS Collada

[Stefano Corazza]

I'm writing this in response to the emails around FBX import topic.
Thank you Tom, Bastien, Toni, Jeffrey and Campbell for your input. 
Bastien I also would like to express my admiration for the reverse 
engineering work you have done so far.

Background: Our company - Mixamo - has worked with FBX SDK since 2008. 
We also help wrote some top industry 3D software FBX exporter (I cant 
mention it). We have extended FBX SDK 2009.3 to support Collada 1.4.1. 
We have created a Blender 2.49 plugin to import Collada 1.4.1 which is 
still the only full feature way to get Collada 1.4.1 into Blender. We 
recently upgraded to FBX SDK 2014 and we are constantly in contact with 
Autodesk FBX group to provide them bug reports/test cases. We are very 
focused on character animation specifically.


Collada:

The current Blender Collada import support is 90% there. We have done 
extensive tests with characters, rigs and animations.
Most of the assets work just fine, the only gap at the moment is on the 
support for multiple meshes/bind poses, here a test case we put together:

/https://developer.blender.org/file/info/PHID-FILE-2kleiwpxahd6to6lac5y/#/


FBX:

The current FBX support is still in its initial stage and full support 
would be a huge project to undertake for a format that is constantly 
changing. The things that are missing are related to animation and 
include skinning, definition of bind poses, layers, etc.  The complexity 
of the FBX format is pretty huge. Reverse engineering mesh import is 
orders of magnitude easier than reverse engineering the animation 
system. FBX SDK had several major changes going from the very stable 
2009.3 into 2011 and  2013 including radical and non backwards 
compatible format changes. For all these reasons and many more I would 
be happy to explain I think FBX is not a viable solution in the short or 
long term for the amount of resources it will take and the intrinsic 
lack of full feature support it will lead to.

Autodesk provides a free tool to convert any FBX into Collada: 
http://usa.autodesk.com/adsk/servlet/pc/item?siteID=123112&id=20481519
They have actually improved Collada support recently.


For the reasons above I would STRONGLY recommend that we fix the current 
Collada importer that is 90% there and keep Blender "clean" instead of 
going for the FBX route.

What do you guys think??
Do you want me to provide a more detailed case?


Thanks!!

Stefano