Hi. I experience some weird results of TSG too. Here's the scene with only one object (valid normal mapping): https://dl.dropboxusercontent.com/u/12634473/img/20140315_tsg_valid.png Here's the scene with 2 objects (invalid normal mapping): https://dl.dropboxusercontent.com/u/12634473/img/20140315_tsg_invalid.png Just adding an object to the scene breaks everything. So I'm more than sure that TSG is buggy.
I will post more results when I advance in my investigation. The complete source code of what I'm doing (depends on my forked osgEffectCompositor): http://goo.gl/O5sAmm Thanks. 2013-10-01 16:06 GMT+07:00 Nicoulaie Vernouillet <[email protected]>: > Hi, robert > > here is the test case. these are just two faces that share the same > normal, but have non-collinear uv coordinates. "with_add.osg" file > accumulates the calculated tangent and binormal, while "withou_add.osg" > crushes the previous calculations. > > I made a little diagram to explain what happens at the shared vertex. > we realize that if the direction of uv is not collinear it overwrites the > information from the previous face. And during rasterization, the linear > interpolation is performed in the middle of the first face and not between > the two side of the edge. > > I hope to be clear enough, feel free to ask me any other information if > this is not enough. > > Thank you! > > Cheers, > Nicoulaie[/img] > > ------------------ > Read this topic online here: > http://forum.openscenegraph.org/viewtopic.php?p=56553#56553 > > > > > _______________________________________________ > osg-submissions mailing list > [email protected] > > http://lists.openscenegraph.org/listinfo.cgi/osg-submissions-openscenegraph.org > >
_______________________________________________ osg-submissions mailing list [email protected] http://lists.openscenegraph.org/listinfo.cgi/osg-submissions-openscenegraph.org
