How to control the char number of output mesh?

[KickRooster]

The image above is my unwrapped uv's world position which I used to bake lightmap.The mesh is the gun from the famous PBR assets. I hope to decrease output chart number, but I don't know which paramaeter can do this.
Thanks in advance.

[castano]

It looks as if your mesh was disconnected. The atlas tool is not able to merge disconnected mesh pieces. It expects you to provide a mesh where each connected component has shared vertices. You can join the vertices trivially if you know the mesh is 2-manifold, otherwise it's much more tricky and I recommend exporting the topology information from the authoring tool. Hope this helps!

[castano]

Also, can you point to the asset you are using? I can take a look at it and check if this is indeed the problem.

[KickRooster]

Yes, my testing mesh was disconnected, I have given up all optimization when imported it.The indices have the same count as the vertex, they are all 100629.My testing mesh was downloaded from http://artisaverb.info/PBT.html, you can download it at the bottom of the web page : ).
And then since I have got the reason, I resume vertex merge optimization when imported.Too sad atlas_generate() will access invalid memory address so to throw a exception for this testing mesh.I use assimp to load mesh.I will debug into it to get why it can access invalid memory address, I think it's the last step before I inject thekla_atlas into our baking engine.
Thank you very much for your help.

[KickRooster]

And, I initialized the input mesh as the following code snippet, although I don't know what's the meaning of first_colocal.

[castano]

The first colocal index is used to determine the mesh connectivity. In a mesh with vertices that have shared positions, but different normals or texture coordinates, you need to know if those vertices are connected, or separated. The first colocal index allows you to do that. If two vertices have the same index, it means they are connected. In some cases you can assume that all vertices that share the same position are connected, in that case, you can set the colocal index to the first index of those vertices. However, in some cases, that may result in a non-manifold mesh. Ideally, you should use connectivity information from the DCC app or authoring tool to initialize the first colocal index in a way that produces valid manifold meshes. For example, traversing the one-ring, and setting the index of all vertices in the one-ring. Hope that makes sense!

…

On Thu, May 24, 2018 at 1:26 AM, 陈默 ***@***.***> wrote: And, I initialized the input mesh as the following code snippet, although what's the meaning of first_colocal. [image: 3333] <https://user-images.githubusercontent.com/8615289/40473578-232f6418-5f6f-11e8-834e-f2efcdef75e3.png> — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#18 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAhcM2dWCfJYNWOoSltQG_rKabPd-9Ckks5t1m7RgaJpZM4UJ59A> .

-- Ignacio Castaño castano@thekla.com

[KickRooster]

I have understood the meaning of the first colocal index by your replay.
By using it marking out disconnected edge so I can split mesh into different charts. I will take some time to record vertex merge info when import it to generate the first colocal index info to generate lightmap uv further.
And then I will go to resolve the "non-manifold mesh" problem.Because I have never met it before. : ).

[castano]

You may not have to worry about non-manifold meshes. In many cases meshes are already split into groups and each group is manifold, so as long as you only connect/merge vertices within the same group you should be fine. Anyway, good luck with your project!

[KickRooster]

Hi, I'm back. Thanks for your prompt about non-manifold meshes first, although I have not met it till now. : )
I have been working for filling index info into the first colocal index for these days, and now it works for one of my testing mesh.It looks like this, wonderful!

But there is a mesh making my program crashed when atlas_generate() was called. Would you help me to have a look at whether it crashed on your environment? : ) Here is the compressed mesh file:
Cerberus_LP.zip
Thanks in advance.