Manually upgrading Three.js JSON geometry files from r42 to r55 - three.js

I have some JSON files that define geometries, as required by r42 of Three.js's JSONLoader class. The files are version 2 of the format.
Is it feasible to manually update these files between versions?
The first obvious difference is that the file is no longer JavaScript, but actual JSON. This was easy to correct for. However the new format has a metadata section, and thirty minutes of experimentation with is isn't getting me anywhere.
I'm seeing problems because the Geometry object's material properties has an empty materials array. The resulting geometry has multiple parts with different materials, as seen in r42.
Does anyone know how I might manually tweak these .json files to work with r55?

I have no complete knowledge how the internals of the format have changed, but here's a couple of hints:
If you have the source object, the best way would be to re-export/convert. There should now also be more converters and exporters to use if your source format is obscure. If the source is unknown, some Googling might be worth it.
metadata section doesn't matter, it isn't used for anything in the loader.
There is no more Geometry.materials. Instead, JSONLoader's callback returns the loaded materials as a separate parameter to the callback. See Migration Guide (r52->r53) for more on that. In fact, the loader interface has changed also in r46.
git log of some example model files (searching for changes there could be your way forward if you really need to manually migrate) suggests that there might be e.g. UV flipping which would be difficult to fix manually but could be worked around in code. But first you'd need to get something to display on screen.

Try dragging the file into the editor, then do File/Export Geometry.

Here is a fix for drag and drop into the editor. Add this code before the drop event in index.html. I tested this in Chrome ( 24.0.1312.56 ), Safari and Firefox in OSX.
document.addEventListener("dragover", function(e) {
e.preventDefault();
});

#mrdoob's answer worked after a few patches to the editor (here and here.)
However it's worth noting that upgrading the files by hand in a text editor is possible, especially if you don't have any UV coords.
Earlier versions did not use JSON -- they were JavaScript files. The conversion is straightforward.
You can either ignore the metadata section, or else port it from the comments into an object.
If you do have UV coords, then they must be mapped differently (I believe an axis is flipped)

Related

Three.js calling clipAction.play() makes animated objects vanish

In Three.js, Calling action.play() makes objects just vanish, without any error or warning on the console.
I use THREE.ObjectLoader to load a JSON file created in blender. The srt (position/scale/quaternion) animation is in the generated file. As are the morphtargets. To optimise filesize I animated the srt as a series of null objects. The morphtargets tracks are in the main object, which I clone 5 times to build the characters (balloons to be exact).
I previously did extensive testing to introduce shape/morph animation. After being succesfull I finalised all the animations. Only to be trumped by the disappearing models. The srt (position/scale/quaternion) animation was working fine before. But after refactoring the code, to be less spagettied, upon calling action.play(). The objects just vanish, exactly then. Echoeing the mixers and the array containing the clips, everything looks correct (ie I see the tracks, the names are right etc). Also examining the newly generated JSON, it seems the same and correct (also I have not changed the SRT animations, only introduced shapeanimation)
So I am lost, and think this looks more and more like a bug. From previous experience I do know it works (or has worked).
I created a jsfiddle: https://jsfiddle.net/oompol/3ya6sqed/
[edit] I turned on the action.play and call the function from the link in the div [/edit] please note I commented out calling action.play(). So you see the load and init work. See the function listed below
function playScene(scene) {
for (parentName in srtMixers) {
var clpName = "balloon1_fly";
var clp = THREE.AnimationClip.findByName(animLib, clpName);
var action = srtMixers[parentName].clipAction(clp);
action.clampWhenFinished = true;
console.log("playScene:", clpName, clp, parentName, srtMixers);
//this is when the problem happens
action.play();
}
}
This is the JSON I am loading:
https://rawgit.com/bakajin/2e3d2f6a722103ed4aefd76f6250ec08/raw/28cad35c20060d478499c0cd40a2753611993720/oomp-scene_balloons-oomp-6.9.4.json
Ok,
there was something very wrong with the scaling indeed.
The io_three JSON exporter for Blender (r87 dev) writes incorrect matrix transformation data in the geometry object (really tiny scaling values). The animation track with the scaling keys were correctly written as 1,1,1. So all the objects just scaled out of view immediately.
Hard to see because the geometry has no separate scaling value but a matrix. Seems to happen when you set "Scene" to true on export.
Worked around the problem by entering the scaling value in the keyframe tracks. But this will only work if you have no scaling animation (so the keys are all one).
Meanwhile I have extensively edited the JSON by hand. Because this is not the only incorrect data. The formatting of the animation object is also wrong. The durations for the morphTargetInfluence Keys is also incorrect. The formatting of these keys is also not always correct.
Hope this helps some other ppl

3D model scn file does not rotate or scale in AR kit

I started a new Xcode project with the ARKit template and simply replaced the "ship.scn" with my "test.scn" filename asset. The object is about 16.5mm wide and 4.8mm tall. The ship worked fine of course, but my test object that reads "test" does not rotate as I move around it, or scale when I move towards or away from it, yet it does track in one location.
I compared the ship and test attribute panels, and I can't find anything that is different about them, except that the ship is textured and my test text is not. What is inherently special about scn objects that would make them behave correctly in ARKit besides their size? I've read through the documentation about anchoring, but it seems like I wouldn't have to do this in code if it's already a scn object.
In case anyone wants to test the file I'm using in the ARKit template to see how it's behaving, the file is here: https://ufile.io/ey49t
I’m answering the question because I think it could help others that are making a similar mistake, but it can be closed if it doesn’t make sense.
The problem is that the size was much too large where I could not see scaling or rotation no matter how much I moved around the room. Compare the scaled size in the last attribute panel on the ship model - not its actual size. Then get the size of your own model scaled down enough so that it is actually resonable.

Three JS animation in Editor possible?

I have been able to get almost everything I need into the editor, But cannot figure out how to get a collada model that was imported to play it's keyframes.
Since I am not doing loader.load (collada) blah blah like in the examples. I cannot figure out how to get the animations. Animations is always undefined, but I know keyframes exist. Any ideas?
as far as i know, animations in the editor are not fully implemented yet.
the corresponding code in the viewport is commented out (line564+)
i think your best shot is to modify the editor (and eventually create a PR) or go back to coding everything.
the editor in its current form only shows the basic functions but the code is quite well structured to allow implementing custom features.
A quick update that animations in the Three.js Editor are now supported. Below is a gif from the linked PR demonstrating usage.

Inconsistent transparency, looking like render order issue

I am using r82.
I have a mesh with multiple materials. I can change their opacity just fine, but how they are rendered is what I would call "splotchy". I have been using ThreeJs for a while, and EDIT: was able to get the transparency working in a past version (r67) with the same model in a significantly more consistent way. So I was wondering if there is something that now I need to set that I didn't need to set before or if I am just overlooking something. Upon revisiting my older code and testing it again, I found that the same transparency issues were present. It was simply a matter of there not being as obvious "splotches" (and not testing enough, I'm sure). Here is a screenshot.
Here are a few more pictures I took that highlight the issue a bit better. I have the outside wall in a light grey and the floors a dark grey in the model and can toggle the outside walls to be visible or not. In these pictures I have one face of the outside wall purple and a face of the floor in the room on the other side of the wall green.
Based on the angle of the camera, it makes part of the green floor face invisible even though there is only one face between the camera and it.
The materials are all double sided already and there is no sign of this until the transparency is on. I found a similar question that suggested changing the mesh.setFaceCulling (or something similar) but that seemed to be from an older version and wasn't in r82.
Thanks for any help in advance!
EDIT:
I started looking into the old version of threeJS and the current version's source code to see what is done differently regarding transparency. I found transparentObjects, which is an array of the objects (I believe faces) that are going to be rendered and are in sorted based on "reversePainterSortStable". There is another list of objects (I believe for the materials objects, maybe?) called opaqueObjects that uses "painterSortStable". So to see if changing the sort order would change the outcome of how things are looking when transparent I changed it so that transparentObjects got sorted by "painterSortStable" and it did change how things showed up significantly (granted it didn't fix my problem since it just removed some problem spots and created new ones).
So the short version, it looks like it is an issue with the renderOrder of the faces.
That being said, I tried finding how the r67 version of the code handled the "renderOrder" of the faces since it wasn't something that (to my knowledge) could be set in that version and just did it automatically. But I have had no such luck tracking down how it was done as of yet.
So I see two possibilities. 1) find out how the past version correctly did transparency (at least for my purposes) and change the logic in the current version to use that. Or 2) find how to properly set the renderOrder of the faces based on the camera position in the scene. Will look into the second option first, but figured it would be good to document this for others looking to help answer or that have a similar problem.
EDIT 2:
So digging through the source code for threeJs and noticed something about the transparentObjects array I mentioned in the previous edit. The first, that I cannot for the life of me figure out how it gets populated since it doesn't seem like it is added to anywhere in the code. The second is that I think it is being populated with duplicates of the entire building object/mesh (see screenshot below).
The z indexes all seem to be the same. as well as the ids and the objects are all of type "mesh" (of the ones I looked through, granted, since there are a few thousand). So I was going to figure out why its adding what is being added to the array, but that is when I stumbled across the issue of not finding where in the code that the transparentObjects array actually get populated.
EDIT 3:
WestLangley, I tried setting the depth test for the outer wall material to false and got this. Like I said in my response comment, even if it did work it wouldn't fix the issues experienced with the camera inside the building, but wanted to follow up none the less (see snapshot below).

Overidding imshow

When I use the following,
imshow(imread('image1.jpg'));
imshow(imread('image2.jpg'));
imshow(imread('image3.jpg'));
imshow(imread('image4.jpg'));
imshow(imread('image5.jpg'));
imshow(imread('image6.jpg'));
I got only image named image6.jpg in the output figure.
There is also an option figure,imshow(...); to view all the images each in new window.
But writing figure in each line where I need to view the image is a repeated and tedious process. Is there any other solution to get the same output as with figure,imshow(..);
without using figure function.
I mainly put on this question because while programming a lot somewhere we forget to use the figure function and so the image that we need to view wont be visible. It would have been overwritten by other image. So provide me some solution.
I ask this only for simplicity in writing the code. So if there is any solution, please mention.
Thanks in advance.
I'm not sure, but I don't think that there's a workaround to that. MATLAB basically changes the current figure handle to that of a new image when you use imshow. One thing you can do however is to make a copy of imshow in your local directory and edit it accordingly to make your own UDF.
What I would recommend however (so as to preserve functionality across systems) is that you open your code is an editor and replace all imshows with figure, imshow. This should be easy enough and it'll be easy to revert back as well.

Resources