OutOfTheMemory, estimated .glb file size

[dimiitpk]

Hi @ThomasGorisse,

I'm new to sceneform and AR, we've been building a project for a while and .glb files with 10-30 MB worked well, but now with a 400 MB file I get an exception:

Unable to load Renderable registryId='model_new.glb'
    java.util.concurrent.CompletionException: java.lang.OutOfMemoryError: Failed to allocate a 536870928 byte allocation with 6275072 free bytes and 251MB until OOM, target footprint 279907752, growth limit 536870912
        at java.util.concurrent.CompletableFuture.encodeThrowable(CompletableFuture.java:278)
        at java.util.concurrent.CompletableFuture.completeThrowable(CompletableFuture.java:284)
        at java.util.concurrent.CompletableFuture$AsyncSupply.run(CompletableFuture.java:1629)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1167)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:641)
        at java.lang.Thread.run(Thread.java:919)
     Caused by: java.lang.OutOfMemoryError: Failed to allocate a 536870928 byte allocation with 6275072 free bytes and 251MB until OOM, target footprint 279907752, growth limit 536870912
        at java.util.Arrays.copyOf(Arrays.java:3161)
        at java.io.ByteArrayOutputStream.grow(ByteArrayOutputStream.java:118)
        at java.io.ByteArrayOutputStream.ensureCapacity(ByteArrayOutputStream.java:93)
        at java.io.ByteArrayOutputStream.write(ByteArrayOutputStream.java:153)
        at com.google.ar.sceneform.utilities.SceneformBufferUtils.copy(SceneformBufferUtils.java:75)
        at com.google.ar.sceneform.utilities.SceneformBufferUtils.inputStreamToByteArray(SceneformBufferUtils.java:127)
        at com.google.ar.sceneform.utilities.SceneformBufferUtils.inputStreamCallableToByteArray(SceneformBufferUtils.java:119)
        at com.google.ar.sceneform.rendering.LoadRenderableFromFilamentGltfTask.lambda$downloadAndProcessRenderable$1(LoadRenderableFromFilamentGltfTask.java:52)
        at com.google.ar.sceneform.rendering.-$$Lambda$LoadRenderableFromFilamentGltfTask$U3-Xk_AQ2v3ZU11qqmGEFQQUnBM.get(Unknown Source:2)
        at java.util.concurrent.CompletableFuture$AsyncSupply.run(CompletableFuture.java:1627)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1167) 
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:641) 
        at java.lang.Thread.run(Thread.java:919) 

My question is, is there some recommended file size for this SDK to work fine, or am I making a mistake somewhere?

[devbridie]

The amount of memory varies per device and operating environment, but a 400mb file probably shouldn't be loaded into memory under any circumstances. Check out this section on Android Developers to query memory info, but in this case, you should be looking at lowering file size (e.g. reduce polys / vertices, lower res textures); a 400mb GLB is absolutely huge and probably wouldn't render quickly even if you did manage to load it.

[ThomasGorisse]

@devbridie is right.

For sure you will have to reduce your model size.
Try to lower it down at about 20Mb and you should be able to use it on high memory phones.
There are a lot of ways to do it using Blender.

I personally made a topographical drone scanned monument going down from 1Gb to 15Mb.

Everything will depend on your model source (CAD/Topographic/...) and your model destination (preview or real technical details needed)

First thing you want to do is having a look at the textures used by your model.
In case of topographic datas coming from 3D scanners, those textures might be very high images sources you can reduce using an image editor (GIMP/Photoshop) or replace them with material based values (Filament constantly integrate new material extensions)

Secondly, one case I met was a 3D source coming from factory machinery complex plumbing pipes builders datas.
In case of CAO datas, no textures are used since their goal is only to have a micron precision for every object part (including not visible ones)
In this case, you have to use some Blender modifiers to reduce the face count (low poly)
One of the most popular/used for this case is Decimate

For an ARCore/Filament context (Sceneform) try to adjust the ratio until you got about 100 000 faces max. It's a lot comparing to other mobile AR/3D rendering engines limit (Unity not included from what I know)
If your model doesn't have to much spherical parts, the rendering quality will not be to much impacted.

There are a lot of other things you can do inside Blender to reduce the model size.
You can have a look at blender tutorials or in here and Filament repo like here:

• GLTF not loaded properly #103
• Filament logo 3D Model google/filament#4010

Last thing but not least make sure to check Draco compression when exporting from Blender since Filament handle it:

If you still don't have an acceptable result with all this, you can share your model here and we will try to see what we can do.

[dimiitpk]

Thank you very much for the quick response, we were able to launch the model by reducing its size.