Spark v2 fails to load plain CloudCompare binary PLY point cloud (Missing scale_0 property)

[Huciko]

Spark v2 appears to fail when loading a plain x/y/z/r/g/b CloudCompare-exported binary PLY point cloud.

The same type of file was loading successfully with the previous Spark version I was using, @sparkjsdev/spark@0.1.10. After upgrading to @sparkjsdev/spark@2.0.0, Spark now appears to treat this plain point-cloud PLY as a Gaussian Splat PLY and expects Gaussian-specific properties such as scale_0.

This seems inconsistent with the Spark v2 documentation, which says .ply loading supports original gsplat PLY, compressed SuperSplat/gsplat PLY variants, and plain x/y/z/r/g/b point clouds, with .ply / .spz auto-detected from file contents.

Docs reference: https://sparkjs.dev/docs/loading-splats

Environment
@sparkjsdev/spark: 2.0.0
three: tested in my app with 0.184.0
App: Vite + TypeScript + Three.js

PLY file details
The file is a CloudCompare-exported point-cloud PLY.

Original header:
ply
format binary_little_endian 1.0
comment Created by CloudCompare v2.13.0 (Kharkiv - Feb 14 2024)
comment Created 2024-03-29T00:12:28
obj_info Generated by CloudCompare!
element vertex 279910
property float x
property float y
property float z
property uchar red
property uchar green
property uchar blue
end_header

There are no Gaussian Splat properties such as:
scale_0
scale_1
scale_2
rot_0
rot_1
rot_2
rot_3
opacity

This is expected because it is a plain point-cloud PLY with positions and RGB colors.

What I tried

1. Direct SplatMesh URL load, no fileType
   const tree = new SplatMesh({
   url: './images/tree.ply',
   });

scene.add(tree);

Result:

Invalid PLY file
2. Removed CloudCompare obj_info line

Because Spark first complained about:

Unsupported PLY header line: obj_info Generated by CloudCompare!

I created a copy of the binary PLY with only this header line removed:

obj_info Generated by CloudCompare!

The binary vertex body was left untouched.

Then I tested:

const tree = new SplatMesh({
url: './images/Tree-no-obj-info.ply',
});

scene.add(tree);

Result:

Missing scale_0 property
3. Direct SplatLoader, no explicit file type
const loader = new SplatLoader();

const loadedSplats = await loader.loadAsync('./images/Tree-no-obj-info.ply');

const tree = new SplatMesh({
packedSplats: loadedSplats as PackedSplats,
});

scene.add(tree);

Result:

Missing scale_0 property
4. Ensured no explicit SplatFileType.PLY

I also changed my own loader mapping so that .ply is not passed as an explicit fileType, allowing Spark to auto-detect:

function getExplicitSparkFileType(format?: SplatFormat): SplatFileType | undefined {
switch (format) {
case 'ply':
// Let Spark auto-detect PLY variants from file contents.
return undefined;

    case 'spz':
        return SplatFileType.SPZ;

    case 'ksplat':
        return SplatFileType.KSPLAT;

    case 'splat':
        return SplatFileType.SPLAT;

    default:
        return undefined;
}

}

So for .ply, Spark receives:

new SplatMesh({ url });

or:

new PackedSplats({ url });

not:

new SplatMesh({
url,
fileType: SplatFileType.PLY,
});

The issue still occurs.

Minimal reproduction

This minimal Spark setup works with .spz files:

<style> body { margin: 0; } </style> <script type="importmap"> { "imports": { "three": "https://cdnjs.cloudflare.com/ajax/libs/three.js/0.180.0/three.module.js", "@sparkjsdev/spark": "https://sparkjs.dev/releases/spark/2.0.0/spark.module.js" } } </script> <script type="module"> import * as THREE from "three"; import { SparkRenderer, SplatMesh } from "@sparkjsdev/spark"; const scene = new THREE.Scene(); const camera = new THREE.PerspectiveCamera( 60, window.innerWidth / window.innerHeight, 0.01, 1000 ); const renderer = new THREE.WebGLRenderer(); renderer.setSize(window.innerWidth, window.innerHeight); document.body.appendChild(renderer.domElement); const spark = new SparkRenderer({ renderer }); scene.add(spark); // This works: const spz = new SplatMesh({ url: "https://sparkjs.dev/assets/splats/butterfly.spz", }); spz.position.set(0, 0, -3); scene.add(spz); renderer.setAnimationLoop(() => { renderer.render(scene, camera); spz.rotation.y += 0.01; }); </script>

But replacing the SPZ load with the plain CloudCompare PLY fails:

const tree = new SplatMesh({
url: './Tree-no-obj-info.ply',
});

tree.position.set(0, 0, -3);
scene.add(tree);

Error:

Missing scale_0 property
Expected behavior

A plain point-cloud PLY with:

property float x
property float y
property float z
property uchar red
property uchar green
property uchar blue

should load as a point-cloud splat source, based on the Spark v2 documentation.

Spark should not require Gaussian Splat fields such as scale_0 for this plain point-cloud PLY path.

Actual behavior

Spark v2 appears to treat the file as a Gaussian Splat PLY and throws:

Missing scale_0 property

Before removing the CloudCompare metadata line, it also throws:

Unsupported PLY header line: obj_info Generated by CloudCompare!

Questions
Is plain binary little-endian x/y/z/r/g/b PLY currently supported in Spark v2?
Should obj_info PLY header lines be ignored as valid PLY metadata?
Is this a regression from Spark 0.1.x?
Is there a different loading option needed for plain point-cloud PLY files in Spark v2?
If plain point-cloud PLY support is intended, could the loader auto-detect this file as a point-cloud PLY instead of a Gaussian Splat PLY?
Why this matters

I am building a Three.js/Spark-based viewer that needs to support multiple splat/point-cloud formats:

.spz
.splat
.ksplat
Gaussian Splat .ply
plain point-cloud .ply from tools like CloudCompare

SPZ works correctly in Spark v2, but this plain CloudCompare PLY no longer loads after upgrading to Spark v2.