Feat/native ply io

[madsendennis]

Native scala implementation to read/write the PLY format. The implementation is ~20 times faster to write a PLY file and ~3 times faster to read.

Currently is still missing support for TexturedFile.

This is the first part of removing the VTK dependency in Scalismo, https://github.com/unibas-gravis/scalismo/issues/415

[Andreas-Forster]

There is an implementation in scalismo.faces for reading and writing plys. How does this implementation relates to the other one? Is it completely independent?

[madsendennis]

There is an implementation in scalismo.faces for reading and writing plys. How does this implementation relates to the other one? Is it completely independent?

Ohh damn. Should have looked in there first. This one is completely independent. Maybe we should just copy over the scalismo faces version instead. I found a few java implementations, but they all missed a few things, so ended up creating one from scratch.

[madsendennis]

The PLYMesh write/reader in scalismo.faces would also work. It has a lot more structure to it. And this library has been tested a lot more. The question is how to structure it such that we only move all the parts that has with the format supported in scalismo. And, the implementation is rather slow: half the speed on reading and ~15 times slower for writing (compared to the one suggested in the MR).

[Andreas-Forster]

Regarding the two options:

• Using your fast and simple here and keep the feature rich in scalismo-faces
• Move the feature rich but a bit slower here. I have no clear preference. Maybe I would go with a simple fast version in the core scalismo and have the feature rich one as additional module available, as it is now with scalismo-faces.

[madsendennis]

I'll have a closer look at how the file is read/written and see if we can do some improvement on the scalismo-faces version. I think it would look nicer to have an extension of the base-ply functionality in scalismo-faces. But speed is key for me - so only if it's not slower. For now I put the MR in draft mode.

[madsendennis]

Some more profiling between my simple method found in the PLY.scala file and the ported scalismo-faces ply reader found under the io/ply folder:

Facemesh with 53.466 vertices and 106014 triangles: Writing test (16 times faster)

• PLY: 0.051 seconds
• scalismo-faces: 0.824 seconds

Reading test (5.5 times faster)

• PLY: 0.07 seconds
• scalismo-faces: 0.385 seconds

For a larger mesh with 575.781 vertices and 1.147.466 triangles Writing test (40 times faster)

• PLY: 0.239 seconds
• scalismo-faces: 8.034 seconds

Reading test (3.3 times faster)

• PLY: 0.801 seconds
• scalismo-faces: 2.773 seconds

I did try to swap the reader in scalismo-faces with the java ByteBuffer as I use, but at best I could make it double the speed. There is a lot of inefficient structure in the faces implementation with large mutable data structures that are appended to. I did try to change some of these to predefined Arrays, but also with limited success.

So I'm more in favour of keeping a simple reader in scalismo and having a more advanced (also for texture meshes) in scalismo-faces.

But maybe someone else can see how we can somehow easily make a merge of the two?

[Andreas-Forster]

I would be open to merge the fast PLY IO and later check if we can speedup scalismo-faces. As long as they can read each others files for those parts that both support, this is the best option with the least overhead.

[madsendennis]

I updated the ply reader to follow the structure of the stl reader to be more coherent. Hopefully, it will be easier to read now that it is split across multiple files. I also added a more general way to read properties from the ply file. Now it reads all kinds of properties (also texture if it is there), but it only makes use of the items requested, i.e. only vertex and faces if requesting to read a TriangleMesh.