SkinnedMesh: support weight normalization for BufferGeometry

[WestLangley]

Fixes #7111 and #7197.

/ping @titansoftime Please have a look.

[mrdoob]

setXYZW() maybe?

[WestLangley]

Oops... Good eyes! : - )

[titansoftime]

It seems the // do nothing are still necessary. Without sw.set(1) and vec.set(1) I get some fairly odd visuals:

I will get a fiddle of this up soon.

With vec.set(1):

Without:

[WestLangley]

Please see if you can track down what is going on with your models via the debugger.

Vector4.set() takes 4 args, not 1.

[titansoftime]

Here's the fiddle: http://jsfiddle.net/cndoe3re/

The set(1) was just what I took from the original function.

THREE.SkinnedMesh.prototype.normalizeSkinWeights = function () {

    if ( this.geometry instanceof THREE.Geometry ) {

        for ( var i = 0; i < this.geometry.skinIndices.length; i ++ ) {

            var sw = this.geometry.skinWeights[ i ];

            var scale = 1.0 / sw.lengthManhattan();

            if ( scale !== Infinity ) {

                sw.multiplyScalar( scale );

            } else {

                sw.set( 1 ); // this will be normalized by the shader anyway

            }

        }

    } else {

        // skinning weights assumed to be normalized for THREE.BufferGeometry

    }

};

EDIT* More info:

This url: http://www.titansoftime.com/webgl/Three74devb.full.js is used in http://jsfiddle.net/he21xL1p/ with the following function:

THREE.SkinnedMesh.prototype.normalizeSkinWeights = function () {

    if ( this.geometry instanceof THREE.Geometry ) {

        for ( var i = 0; i < this.geometry.skinWeights.length; i ++ ) {

            var sw = this.geometry.skinWeights[ i ];

            var scale = 1.0 / sw.lengthManhattan();

            if ( scale !== Infinity ) {

                sw.multiplyScalar( scale );

            } else {

                sw.set( 1 );

            }

        }

    } else if ( this.geometry instanceof THREE.BufferGeometry ) {

        var vec = new THREE.Vector4();

        var skinWeight = this.geometry.attributes.skinWeight;

        for ( var i = 0; i < skinWeight.count; i ++ ) {

            vec.x = skinWeight.getX( i );
            vec.y = skinWeight.getY( i );
            vec.z = skinWeight.getZ( i );
            vec.w = skinWeight.getW( i );

            var scale = 1.0 / vec.lengthManhattan();

            if ( scale !== Infinity ) {

                vec.multiplyScalar( scale );

            } else {

                vec.set( 1 );

            }

            skinWeight.setXYZ( i, vec.x, vec.y, vec.z, vec.w )

        }

    }

};

[WestLangley]

Like I said, Vector4.set( 1 ) is not a proper use of the method. Please see if you can figure out what is going on with your models. I expect you have all-zero weights in places.

[titansoftime]

I am not the one who originally put in set(1), I'm not sure who wrote that, I simply copied it to your BufferGeometry section of the function.

I've had this model since r58 and it has been fine until now. So I highly doubt it's the model. I am not a modeler or nor an animator so any investigation on my part would be useless. I can tell you that the weights are not all 0.

http://www.titansoftime.com/utils.php?task=getModel&id=121

http://jsfiddle.net/he21xL1p/ works with the set(1) and http://jsfiddle.net/cndoe3re/ does not work without it. The function originally had the set(1), it's not of my doing, but it seems to work just fine. It's absence yields breakage. Perhaps the original author of the function THREE.SkinnedMesh.prototype.normalizeSkinWeights could shed some light on why they used sw.set( 1 ). sw is a Vector4 of course.

I'm just saying that what he/she did appears to have a reason though it's missing 3 args. Evidence being it works with it and not without it. I'm just the messenger =]

[titansoftime]

lol @ // do something reasonable. Works great!

Something was going on other than skin weights of 0 though. You can see here that the weights are not 0:

http://www.titansoftime.com/utils.php?task=getModel&id=121

Regardless, all seems well. Thank you for your efforts =]

[WestLangley]

I expect you have all-zero weights in places.

And you do. Some vertices have all 4 weights equal to zero.

Also, I do not think you give yourself enough credit. I would not have asked for your help if I felt investigation on your part "would be useless" or you were "just a messenger".

[titansoftime]

Some vertices have all 4 weights equal to zero.

Ah, I thought you were saying all weight at 0. Gotcha. A big "duh" on that lol. I assumed js worked like php and would halt on division by 0. Just when you think you know javascript...

Also, I do not think you give yourself enough credit.

That means a lot, thank you. I feel like you guys are beings from the 5th dimension when it comes to this kind of stuff though =]

[mrdoob]

Many thanks guys!

[MasterJames]

As a side note every single example anyone has posted since well before I pointed out a month or two ago - continue to show up black ( on Samsung Tab3 chrome current browser ), but I thought we already know that and have addressed it so it must be that people are using older versions or something. Is this expected or did that fix not go through yet?

[titansoftime]

The model is black on my tablet due to the fact that I am loading a dds texture which is not supported on my samsung galaxy 4. I suppose I should have just used a png for examples sake.

Or, are you saying the entire scene is black?

[MasterJames]

Oh I see that explains this one. No it was just the model. The problem before was in the lighting and/or shader and nothing to do with models. I would have expected some visible shading even if the texture didn't load. I see a flat black palm and solid bright green floor on a white BG.