Migrate to the new animation system

[Trass3r]

Found this old #362 branch and rebased fwiw.

[Trass3r]

I guess this could be a first step towards solving #409 (comment).
AFAICT you'd need to define MorphTargets for the target Geometry instead of custom Poses?

[tonihele]

The jME morph targets work a bit differently. If I remember correctly, they are hardware accelerated. But there was some reason that they don't really work with our way of thinking... Was it that we would then have millions of morph targets per model? And it just didn't work out... Something like this? And the hardware acceleration worked only up until 7 morph targets...

[Trass3r]

Yeah I think it's similar to what I do in blender, 1 morph target per frame and weights like (0,0,0,1,0,0...), so quite a few.
But maybe MorphTrack could otherwise be shoe-horned into doing what we need.

https://github.com/jMonkeyEngine/jmonkeyengine/blob/af375391ea385a061444ca03262e1a4720c3274d/jme3-core/src/main/java/com/jme3/anim/tween/action/ClipAction.java#L49-L53

https://github.com/stephengold/Maud/blob/5ebfd97c3014b616ef520e0038640b9760020e30/src/main/java/maud/model/cgm/LoadedAnimation.java#L612

https://github.com/jMonkeyEngine/jmonkeyengine/blob/master/jme3-core/src/main/java/com/jme3/anim/MorphControl.java#L117

[tonihele]

Yeah, I mean if it works and the performance is ok. Absolutely then yes. I could have just made a quick call back then.

By the way, F5 displays FPS, F6 gives kinda detailed information about the frames being rendered. There can maybe see how the new animations compare to the old, performance wise. Not super accurate but can give some sort of pointer.

[Trass3r]

Thanks!

The problem is that the whole PoseTrack thing is even harder to understand than KMF itself. All those previous and next frame refs in FrameInfo and Pose, all the maps and copy paste calls in:

OpenKeeper/src/toniarts/openkeeper/tools/convert/KmfModelLoader.java

Line 414 in 604a33a

Pose p = new Pose(index + ": " + frame + ", " + entry.getKey().previousPoseFrame + " - " + entry.getKey().nextPoseFrame, frameOffsets.get(frame).get(entry.getKey()).toArray(new Vector3f[frameOffsets.get(frame).get(entry.getKey()).size()]), array);

There also seems to be quite some redundancy:
https://gist.github.com/Trass3r/cdd614ec24c48f94969389492117fa0d#file-piranha_swim-xml-L163

[Trass3r]

Looks like we were already passing through this code, it works cause in getDataAtTime we never set the output weights parameter and then setMorphState is a no-op:

    private void interpolateMorphTrack(double t, MorphTrack track) {
        Geometry target = track.getTarget();
        float[] weights = target.getMorphState();
        track.getDataAtTime(t, weights);
        target.setMorphState(weights);

Btw the shader is indeed limited to a handful of morph targets to blend but you don't really need that many, and the system also falls back to CPU morph target merging as needed:
https://github.com/jMonkeyEngine/jmonkeyengine/blob/af375391ea385a061444ca03262e1a4720c3274d/jme3-core/src/main/java/com/jme3/anim/MorphControl.java#L157-L158

[Trass3r]

Iirc this fixes loading into Maud, but when playing the animation it just shows the base mesh.

[tonihele]

What is the status of this? I didn't try the code yet, do the different animation modes still work?

[Trass3r]

So far I couldn't spot any regressions.
But please feel free to test it yourself.

[tonihele]

On 3.6 jME there will be native tweens for the deprecated loop modes. So lets look at this then.

[Trass3r]

I see.
https://github.com/jMonkeyEngine/jmonkeyengine/releases/tag/v3.6.0-alpha1

[Trass3r]

Would we need to use AnimLayer then?
jMonkeyEngine/jmonkeyengine#1845

[tonihele]

Hmm, not sure, it was this PR https://github.com/jMonkeyEngine/jmonkeyengine/pull/1849/files.

[Trass3r]

Some enhancement to new animation system by @Ali-RS in jMonkeyEngine/jmonkeyengine#1845

Added a Loop tween to Tweens factory class. Supports looping by count… by @Ali-RS in jMonkeyEngine/jmonkeyengine#1846

Added Tweens.cycle() and Tweens.invert() methods by @Ali-RS in jMonkeyEngine/jmonkeyengine#1849

[Trass3r]

It has landed: https://github.com/jMonkeyEngine/jmonkeyengine/releases/tag/v3.6.0-stable

[tonihele]

Yeah, I finally got around to test this. It seems to work.

I tried to use the Tween.cycle in the AnimationLoader rather than tamper with the Global Animation Speed. But that resulted in doors spawning as opened. Well, need to check this at some point.

So yes, fine by me. Ready to merge?