search

LIU-Yuxin / SyncMVD

Official PyTorch & Diffusers implementation of "Text-Guided Texturing by Synchronized Multi-View Diffusion"
MIT License
104 stars 5 forks source link

Hi, met some problems when generating texture for custom meshes. #14

Closed Colezwhy closed 5 days ago

Colezwhy commented 1 month ago

Thank you for your great job! But I met the following problem when trying to load my glb mesh and generate texture, would you give me some suggestions on that? image

Colezwhy commented 1 month ago

Or could you please provide some suggestions on the input mesh format? My original mesh is in ply format, but I converted it into glb format.

Colezwhy commented 1 month ago

Sorry for repetitive bothering. Is it possible that the only input is an .obj file? Also, When changing the text prompt in the demo configs, I don't see any differences between prompts.

Colezwhy commented 1 month ago

textured_views_rgb The generated texture is like this, it seems that this cannot be correct.

LIU-Yuxin commented 1 month ago

The input object is assumed to have only one mesh, and only one material, and a valid UV mapping (enable auto-uv if the UV of original object has overlaps). You can manually cleanup the object in blender. I will also post a blender script for cleanup later.

Colezwhy commented 1 month ago

The input object is assumed to have only one mesh, and only one material, and a valid UV mapping (enable auto-uv if the UV of original object has overlaps). You can manually cleanup the object in blender. I will also post a blender script for cleanup later.

Thank you, but would you please look at the image I shared above? Is the generated texture for the demo 'face' valid?

LIU-Yuxin commented 1 month ago

No, it is not. Are you just running the current program without any change?

Colezwhy commented 1 month ago

No, it is not. Are you just running the current program without any change?

Yes, for sure. And for the two demos they are all like this.

LIU-Yuxin commented 1 month ago

I will check the output when I incorporate some new changes into the code later, I will fix it if I can reproduce the bug. But there is no guarantee that the program will also work on a different system and environment setup.

Colezwhy commented 1 month ago

I will check the output when I incorporate some new changes into the code later, I will fix it if I can reproduce the bug. But there is no guarantee that the program will also work on a different system and environment setup.

Sure, thank you very much. And below is my system and environments. OS: Ubuntu 18.04 GPU platform: RTX3090 24G conda package environment: the same as provided readme, pytorch3d used command 'pip install --no-index --no-cache-dir pytorch3d -f https://dl.fbaipublicfiles.com/pytorch3d/packaging/wheels/py38_cu117_pyt201/download.html'

LIU-Yuxin commented 1 week ago

I did not find any problem causing this issue when I update the code recently. I wonder if it is related to the UV of the object. Maybe you can pull the latest version and check whether you are running with auto UV enabled.

Colezwhy commented 5 days ago

I did not find any problem causing this issue when I update the code recently. I wonder if it is related to the UV of the object. Maybe you can pull the latest version and check whether you are running with auto UV enabled.

Thank you, I have found the problem, it's like after I have cleaned up the huggingface storage and run the script again, it functioned well again. I am closing this. Thank you for that again.