Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Where can I find models of T2I V2 and V3? #311

Open
jakechai opened this issue Sep 21, 2024 · 4 comments
Open

Where can I find models of T2I V2 and V3? #311

jakechai opened this issue Sep 21, 2024 · 4 comments

Comments

@jakechai
Copy link
Contributor

Where can I find models of CRM T2I V2 and V3?
missing

@MrForExample
Copy link
Owner

MrForExample commented Sep 21, 2024

Haha, you are the first to noticed :) that's a PBR textures generation model I'm researching and training myself, and haven't finish it yet, but almost, I may release it after a month or two along with research paper for the whole development process.

Given a 3D mesh and a reference image, it can generate you a 3 texture maps (albedo, metalness, roughness), To show you a few testing results:

cup_render_00001.mp4
car_render_00001.mp4
gun_render_00001.mp4

Cheers, have a good day 👍

@jakechai
Copy link
Contributor Author

jakechai commented Sep 21, 2024

Great! I can't wait to try it. I have some thoughts about your texture projection process. We don't need to make 2nd mesh and texture refinement in 3D software if you could make all these happen.

  1. Can you add an orthographic camera switch for Mesh Orbit Renderer and Explicit Target Color Projection? In some cases, orthographic projection is much better than perspective one. That is a 90 FOV orthographic camera with a Target Distance of 1.0 and a orbit radius of 2.0 based on a 1.0 unit cube, which is the CRM algorithm (the following image). This is based on my experiences on the UE texture projection project.
    crm camera
crm.camera.mp4
  1. Is it possible to bake all the PBR textures such as Albedo, Roughness, and Normal, (Metallic, AO, Cavity, etc.) at the same time? All of them can be converted from the Albedo texture using different methods. One of my favorites is Materialize.
    materialize

  2. Because we need mesh optimization before the texture projection process. Can you bring Mesh Anything the powerful mesh optimization algorithm into the 3D Pack world? This will make the 3D pack a more complete and practical toolkit for mesh reconstruction.

@GeorgeSte
Copy link

Those examples looks awesome! Iv been wondering about the CRM T2I V3 :D Plus the requirements needed. Im guessing its a normal map and single image of the subject? or MV images and MV normals? saw the example workflow having a image bath loaders. Keen to try it out. Also fantastic points from Jake, would be wonderful to have orthographic camera switch for Mesh Orbit Renderer and Explicit Target Color Projection. Had issues where the top of a mesh wouldn't have color projected on its only possible to run images with 0 angle as far as I know.

Exiting development on this all!

@MrForExample
Copy link
Owner

Hi @jakechai

  1. I'll see if I can add that orthographic camera switch for Mesh Orbit Renderer and Explicit Target Color Projection
  2. Right now, the Mesh class can only support single UV and single Albedo, single Metallic (Blue channel) & Roughness (Green channel) map, so as right now, Explicit Target Color Projection node support those three types of maps, I do have plan to extend the Mesh class's functionality more in the future
  3. I have tried Mesh Anything multiple times before, so it's my colleagues, for both v1 and v2, but we all think it doesn't work as robust as advertised, most meshes are kind of unusable, so that is on hold as for now.

Hi @GeorgeSte thanks for the kind words, it uses 6 orthogonal MV normals to condition the shape, and one front view images to condition the appearance, and generate 6X3 MV views for Albedo, Metallic and Roughness respectively

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants