Marmalade SDK (.group.bin) 3D mesh extraction

[IDANAND]

Need help with Marmalade SDK (.group.bin) 3D mesh extraction - Tone Sphere

Hello everyone! I am trying to extract 3D meshes from the mobile game Tone Sphere (Marmalade SDK).

I used a script to split the mono .group.bin file into separate assets. Now I have a 1.4 MB file (dc3c2177) which definitely contains a 3D mesh.

Here is what I found out using HEX and Model Researcher:

Faces / Indices Offset: starts exactly at 0x15C550 (Format: 16-bit Unsigned Short).

Maximum Index: 60288

Vertices: Located right before the faces, but they seem to use fixed-point math (IwFixed / Short_Signed with scaling), which causes a huge mess and wrong scaling when reading via Model Researcher.

UV / Normals: Also packed right before the vertices.

Could anyone help me write a quick Noesis plugin or a Python script to properly parse this mesh? I can provide the sample file! Thanks in advance!

dc3c2177.zip

[shak-otay]

Hi, you missed to tell where the vertices start?

I could only find some strange point cloud:

edit: big parts seem to contain texture data:

Edited yesterday at 09:38 AM by shak-otay

[IDANAND]

Indices Offset: 0x15C550

Indices Type: UInt16 (Short)

Indices Count: 16 026 (5342 triangles)

Max Index Value: 2667 (Which means Vertex Count = 2668)

Vertices Offset: 0x1586C8

Vertex Type: Int16 (Short)

Stride: 6 bytes (Pure X, Y, Z by 2 bytes each)

Coordinate Scale (Fixed-point): Try dividing by 4096.0 or 256.0 (Standard Marmalade SDK shift).

[IDANAND]

I can assume that there are 3 models there at once

 

[shak-otay]

Thanks. Which file is it for? For dc3c2177 your face index parameters would break the file end (15C79F), 0x15C550 + 16026x2 (dec.)= 0x164284

Your vertex offset doesn't show me a decent mesh:

edit: the vertices start at 0x155D38, see one of my next posts

 

Edited yesterday at 01:53 PM by shak-otay

[IDANAND]

thank
For file dc3c2177. Honestly, I don't quite understand how to search for vertices, so I'm just pointing my finger at it. Perhaps I've send the parameters for a different file altogether. I was checking the lesson level file. What I send is the title screen.
And I also know there may be multiple models embedded there. Or it's not a model at all, but a level scene.

And file 8081e087 is just its name, it says title screen there

Edited yesterday at 12:29 PM by IDANAND

[IDANAND]

https://drive.google.com/file/d/1JLP_TtjDKwTE8OJrBkzFC87Lxj9s7vlv/view?usp=drive_link
Here's a link to all the assets used there.
The models are also in group.bin.

 

[shak-otay]

In Backgroundcubes.group.bin I found this but it's hard to extract from the point cloud:

 

[other1]

In dc3c2177 there are multiple meshes, for the last one:
At 1421408, 320 x 3 shorts (position?)
At 1423344, 320 x 3 shorts in range [-4096;4096] (normals?) 
At 1425278, 320 x 2 shorts (UV, divide by 4096)
At 1426576, 384 shorts (indices)

UV looks fine, but mesh is a mess

[shak-otay]

When you use unsigned shorts it looks more like a cube:

edit: another one (I'd create a tool as soon as we get the meshes nicer):

Similar for the background_cubes group.bin:

Edited yesterday at 02:06 PM by shak-otay

[IDANAND]

Here are the ingame screenshots of what these meshes should look like. First one is the single crystal Or a few crystals.

 

[IDANAND]

This is what the levels look like :

[IDANAND]

13 hours ago, other1 said:

In dc3c2177 there are multiple meshes, for the last one:
At 1421408, 320 x 3 shorts (position?)
At 1423344, 320 x 3 shorts in range [-4096;4096] (normals?) 
At 1425278, 320 x 2 shorts (UV, divide by 4096)
At 1426576, 384 shorts (indices)

UV looks fine, but mesh is a mess

Hey! Thanks for the render, but the mesh geometry looks distorted. In the original game, this crystal is a perfectly symmetrical, sharp, elongated octahedron (diamond shape). On your screenshot, it looks flattened and skewed on one side. Look a screenshot.

 

[IDANAND]

13 hours ago, shak-otay said:

In Backgroundcubes.group.bin I found this but it's hard to extract from the point cloud:

 

In this file, the wall should be completely white.