Stability-AI / generative-models

Generative Models by Stability AI
MIT License
24.73k stars 2.75k forks source link

Obtaining meshes/.obj from sv3d? #316

Open FlonneEx opened 8 months ago

FlonneEx commented 8 months ago

The paper makes several references to meshes, rather than just a sequence of images, and the website describing the tech has viewable examples which appear to be .obj files. However, I don't see any option to export an .obj, or any examples, and it isn't obvious how to do so while looking at the scripts.

Am I missing something here? How can I obtain textured 3D objects like the examples?

wulouzhu commented 8 months ago

same question

liu3xing3long commented 8 months ago
hello2mao commented 8 months ago

+1

indussky8 commented 8 months ago

+1

merkol commented 7 months ago

+1

Mohitasudani commented 7 months ago

+1

cycloarcane commented 7 months ago

+1

zhaosheng-thu commented 6 months ago

+1

ivantan-ys commented 6 months ago

From what I understand, the model generates multi-view images from different angles but not the mesh itself (obj, ply files etc.)

This is currently an "unsolved" problem. You can however, theoretically use the output images and feed it through 3D reconstruction software such as Instant NGP, Meshroom or other models that generate mesh from multiple images.

bittergourd1224 commented 6 months ago

From what I understand, the model generates multi-view images from different angles but not the mesh itself (obj, ply files etc.)

This is currently an "unsolved" problem. You can however, theoretically use the output images and feed it through 3D reconstruction software such as Instant NGP, Meshroom or other models that generate mesh from multiple images.

@ivantan-ys However, they introduce a Coarse-to-Fine Training method to generate 3D mesh (compared to original NeRF approach), which is included in the 4th part of their paper, while is completely missing in this repository.

zydmtaichi commented 4 months ago

@ivantan-ys However, they introduce a Coarse-to-Fine Training method to generate 3D mesh (compared to original NeRF approach), which is included in the 4th part of their paper, while is completely missing in this repository. +1, I also want to know the code about their "Coarse-to-Fine Training method" to generate 3D mesh

Jun-Pu commented 1 month ago

Any updates?👀